Generalized Optimal Control of Linear Systems with Distributed Parameters (Applied Optimization)

Generalized Optimal Control of Linear Systems with Distributed Parameters Applied Optimization Volume 69 Series Edito...

Author: S.I. Lyashko

16 downloads 403 Views 6MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Generalized Optimal Control of Linear Systems with Distributed Parameters

Applied Optimization Volume 69

Series Editors: Panos M. Pardalos University of Florida, U.S.A. Donald Hearn University of Florida, U.S.A.

The titles published in this series are listed at the end of this volume.

Generalized Optimal Control of Linear Systems with Distributed Paramaters by

Sergei I. Lyashko Kiev National Shevchenko University, Ukraine

KLUWER ACADEMIC PUBLISHERS NEW YORK, BOSTON, DORDRECHT, LONDON, MOSCOW

eBook ISBN: Print ISBN:

0-306-47571-5 1-4020-0625-X

©2002 Kluwer Academic Publishers New York, Boston, Dordrecht, London, Moscow Print ©2002 Kluwer Academic Publishers Dordrecht All rights reserved No part of this eBook may be reproduced or transmitted in any form or by any means, electronic, mechanical, recording, or otherwise, without written consent from the Publisher Created in the United States of America Visit Kluwer Online at: and Kluwer's eBookstore at:

http://kluweronline.com http://ebooks.kluweronline.com

v

Contents General Preface

ix

Preface

xi

1

1

OPTIMIZATION OF LINEAR SYSTEMS WITH GENERALIZED CONTROL

1

1. FORMULATION OF OPTIMIZATION PROBLEM FOR DISTRIBUTED SYSTEMS AND AUXILIARY STATEMENTS 1 2. EXISTENCE OF OPTIMAL CONTROL 13 2

45

GENERAL PRINCIPLES OF INVESTIGATION OF LINEAR SYSTEMS WITH GENERALIZED CONTROL 1. DIFFERENTIAL PROPERTIES OF PERFORMANCE CRITERION 2. REGULARIZATION OF CONTROL 3

45 45 78 107

NUMERICAL METHODS OF OPTIMIZATION OF LINEAR SYSTEMS WITH GENERALIZED CONTROL 1. PARAMETRIZATION OF CONTROL 2. PULSE OPTIMIZATION PROBLEM 3. ANALOGUE OF THE GRADIENT PROJECTION METHOD 4. ANALOGUE OF THE CONDITIONAL GRADIENT METHOD 5. PROBLEMS OF JOINT OPTIMIZATION AND IDENTIFICATION 6. GENERAL THEOREMS OF GENERALIZED OPTIMAL CONTROL

107 107 112 115 119 123 132

vi 4

PARABOLIC SYSTEMS 1. GENERALIZED SOLVABILITY OF PARABOLIC SYSTEMS 2. ANALOGUE OF GALERKIN METHOD 3. PULSE OPTIMAL CONTROL OF PARABOLIC SYSTEMS 5

PSEUDO-PARABOLIC SYSTEMS 1. GENERALIZED SOLVABILITY OF PSEUDOPARABOLIC SYSTEMS (THE DIRICHLET INITIAL BOUNDARY VALUE PROBLEM) 2. GENERALIZED SOLVABILITY OF PSEUDOPARABOLIC SYSTEMS (THE NEUMANN INITIAL BOUNDARY VALUE PROBLEM) 3. PULSE OPTIMAL CONTROL OF PSEUDOPARABOLIC SYSTEMS (THE DIRICHLET INITIAL BOUNDARY VALUE PROBLEM) 4. PULSE OPTIMAL CONTROL OF PSEUDOPARABOLIC SYSTEMS (THE NEUMANN INITIAL BOUNDARY VALUE PROBLEM) 6

HYPERBOLIC SYSTEMS

137 137 137 145 154 161 161

161

168

173

175 179 179

1. GENERALIZED SOLVABILITY OF HYPERBOLIC SYSTEMS (THE DIRICHLET INITIAL BOUNDARY 179 VALUE PROBLEM) 2. GENERALIZED SOLVABILITY OF HYPERBOLIC SYSTEMS (THE NEUMANN INITIAL BOUNDARY 186 VALUE PROBLEM) 3. ANALOGIES OF GALERKIN METHOD FOR HYPERBOLIC SYTSEMS 192

vii 4. PULSE OPTIMAL CONTROL OF HYPERBOLIC SYSTEMS (THE DIRICHLET INITIAL BOUNDARY VALUE PROBLEM) 204 7

PSEUDO-HYPERBOLIC SYSTEMS

207 207

1. GENERALIZED SOLVABILITY OF PSEUDOHYPERBOLIC SYSTEMS (THE DIRICHLET INITIAL 207 BOUNDARY VALUE PROBLEM) 2. GENERALIZED SOLVABILITY OF PEUDOHYPERBOLIC SYSTEMS (THE DIRICHLET INITIAL BOUNDARY VALUE PROBLEM) 239 3. ANALOGIES OF GALERKIN METHOD FOR PSEUDO-HYPERBOLIC SYSTEMS 283 4. PULSE OPTIMAL CONTROL OF PSEUDOHYPERBOLIC SYSTEMS (THE DIRICHLET INITIAL BOUNDARY VALUE PROBLEM) 308 5. PULSE OPTIMAL CONTROL OF PSEUDOHYPERBOLIC SYSTEMS (THE NEUMANN INITIAL 312 BOUNDARY VALUE PROBLEM) 8

331

SYSTEMS WITH HYPERBOLIC OPERATOR COEFFICIENTS 1. HYPERBOLIC SYSTEM WITH OPERATOR COEFFICIENT 2. GALERKIN METHOD FOR DISTRIBUTED SYSTEMS 3. PULSE OPTIMAL CONTROL OF THE SYSTEMS WITH HYPERBOLIC OPERATOR COEFFICIENT 9

SOBOLEV SYSTEMS

331 331 341 347 349 349

viii 1. EQUATION OF DYNAMICS OF VISCOSE STRATIFICATED LIQUID 2. ONE SOBOLEV PROBLEM 3. PULSE OPTIMAL CONTROL OF THE DYNAMICS EQUATION OF VISCOSE STRATIFIED LIQUID 4. PULSE OPTIMAL CONTROL OF ONE SOBOLEV SYSTEM 10

349 356 361 363 365

CONTROLABILITY OF LINEAR SYSTEMS WITH GENERALIZED CONTROL 1. TRAJECTORY CONTROLABILITY 2. TRAJECTORY-FINAL CONTROLABILITY 3. PULSE CONTROLABILITY OF PARABOLIC SYSTEMS 4. SUBDOMAIN CONTROLABILITY OF PSEUDOPARABOLIC SYSTEMS 11

PERSPECTIVES 1. GENERALIZED SOLVABILITY OF LINEAR SYSTEMS 2. PARAMETRIZATION OF SINGULAR CONTROL IN GENERAL CASE 3. DIFFERENTIAL PROPERTIES OF PERFORMANCE CRITERION IN GENERAL CASE 4. CONVERGENCE OF GRADIENTS METHODS ANALOGUES 5. OPTIMIZATION OF PARABOLIC SYSTEMS WITH GENERALIZED COEFFICIENTS

References

365 365 368 377 384 389 389 389 406 412 420 433 443

ix

General preface Nowadays there are many books on mathematical theory of optimal control of systems with distributed parameters, but as a rule they are devoted to the systems with regular control. The theory of optimization of systems with singular control (including important applied problems such as point, pulse, mobile optimization etc.) is much less elaborated. The book is written by the professor of the Kiev National Taras Shevchenko University S.I.Lyashko. The author made an attempt to create the general theory of optimization of linear systems (both distributed and lumped) with a singular control. This book touches upon wide range of issues such as the solvability of boundary values problems for partial differential equations with generalized right-hand sides, the existence of optimal controls, the necessary conditions of optimality, the controllability of systems, numerical methods of approximation of generalized solutions of initial boundary value problems with generalized data, and numerical methods for approximation of optimal controls. In particular, the problems of optimization of linear systems with lumped controls (pulse, point, pointwise mobile and so on) are investigated in detail. The book undoubtedly will awake the interest of all who is engaged in the theory of optimal control of linear systems and its application in physics, ecology, economy, medicine and other fields. Academician of the National Academy of Sciences of Ukraine

I.V.Sergienko

xi

Preface Intensive development of science and technology put the optimization of various systems in the forefront of applied mathematics and cybernetics. The fundamental results of the optimal control theory were obtained by L.S.Pontryagin, V.P.Boltyansky, R.V.Gamkrelidze, and Y.F.Mischenko [1, 2], A.A.Feldbaum [3], R.Bellmann [4], V.M.Tikhomirov [5], N.N.Krasovsky [6], B.N.Pshenichny [7-9], B.S.Mordukhovich [10], J.Varga [11] and others mathematicians. The theory of control of systems with finite-dimensional phase space was elaborated. But in many technical applications objects have spatial length and its state is described by some classical or non-classical equations of mathematical physics (so called systems with distributed parameters) The investigation of such objects requires considerable generalization of methods of analysis of systems with distributed parameters. The solutions of such problems were obtained by A.Bensoussan [12], B.N.Bublik [13], A.G.Butkovsky [14-16], F.P.Vasilyev [17], A.I.Egorov [18-20], Yu.M.Ermoliev [21-25], V.I.Ivanenko and V.S.Mel'nik [26], J.-L.Lions [27-34], Lurje K.A. [35], A.G.Nakonechny [36], Yu.S.Osipov [37], Yu.I.Samoylenko [16, 38], T.K.Sirazetdinov [39], R.P.Fedorenko [40], V.A.Dykhta [41] etc. Many problems in physics, economics, ecology, medicine etc. are reduced to the problems with state equations, whose right-hand sides include finite order distributions (pulse, point and other controls) [14-16, 42-45]. These problems of singular optimal control yield a series of difficult problems. Although some results were obtained already, but the complete theory was not developed yet. The problem of optimal pulse control of systems with distributed parameters was solved in [6] by introducing into consideration the Stiltjes integral and using the moment L -problem methods. This problem was investigated in [46] as a game one. The papers [47, 48] were devoted to the necessary conditions of optimality in the form of the Pontryagin maximum principle. In [12, 33, 49] the problem of

xii synthesis of the optimal control was reduced to solving of a quasivariational inequality. In the paper [50] the problem of pulse optimization was solved with the help of extension of variational problem and its further analysis in the class of distributions. In [51-53] the problem of point optimal control of systems with distributed parameters was investigated with the help of the semi-group theory. In the paper [20] the pulse position control was obtained for the heat transport equation as a result of solving of some linear problems. The problem of the optimal pulse control was studied in the stochastic formulation in the monograph [12] It should been noted that the introducing of noise into considered systems is equivalent to the regularization and very simplifies the investigation of the optimal controls existence. As regards the necessary conditions of optimality the situation is inverse. The introducing of noise make the formulation of the stochastic maximum principle very complicated. In investigation of controllable systems the problem of its controllability is one of the most important. The problem of controllability of linear systems with lumped parameters, which allow generalized controls, was investigated in [6]. In this paper it was shown that the introducing of such controls does not extend the R.Kalman's conditions of complete controllability. In the case of systems with distributed parameters the state of affairs is much more complicated. It was shown in [14] that the controllability of distributed systems with point controls could be essentially dependent on the numerical nature of the point of the application of control force. Some problems of optimization of systems with generalized controls were considered in the papers [54-61]. Though many problems were solved, numerous urgent problems of generalized control of systems with distributed parameters still either unsolved (for example, problems of pulse controllability and numerical methods) or incompletely investigated (existence of optimal controls, necessary and sufficient conditions of optimality). This monograph describes the investigations in the field of the theory of optimal control of linear systems with distributed parameters with the help of non-linear generalized impacts (including pulse). In

xiii book the theory of optimization of distributed systems elaborated on the basis of a priory inequalities in negative norms is stated. V.P.Didenko [62-64] first obtained the inequalities in negative norms and the method of its proving (integral abc-method) at the early 1970's. This method of investigation of generalized optimization problems is very effective. It allowed to researchers to obtain the results on existence and uniqueness of solutions of initial-boundary value problems, on existence of optimal controls, on necessary and (in some cases) sufficient conditions of controllability in the classes of various generalized impacts. Basing on this method, it is possible to construct numerical methods of generalized optimization (including the methods of solving initial boundary value problems with finite-order distributions in the right-hand sides of its state equations). The book's content is aimed to verify the following theses: 1. In the theory of optimal control of distributed systems we often deal with non-smooth singular controls. That is why we must investigate differential equations in classes of distributions. 2. The Sobolev spaces with negative index and a priori inequalities in negative norms are very suitable for these goals. 3. These methods can be very useful not only in the field of singular control. The first chapters of the book are devoted to the general theory of optimization of linear systems with generalized impacts, for which the a priori inequalities in negative norms hold true. In the following chapters we consider the applications of the general theory to systems described by classical and non-classical equations of mathematical physics, proving. For these systems the validity of these inequalities in negative norms. Also, we construct and investigate numerical methods (analogues of the Galerkin method) to find approximate solutions of initial boundary value problems. Chapter 10 is devoted to the systematic investigation of controllability of systems with pulse, point and similar control. Last chapter contains the recent results and generalizations. We think that the section devoted to generalized solvability of operator

xiv equalities in linear topological spaces is the most interesting. These results undoubtedly will be applied to the control theory. In the book we use the following system of numeration and references. In each section we use a separate numeration of formulae, theorems and definitions. In the frame of single chapter (section) a number of this chapter (section) in references may be omitted. For example, Theorem 1 means the first theorem of the current section in the current chapter, Lemma 1.2 is the second lemma of the first section in the current chapter, (1.2.3) is the third formula of the second section of the first chapter and so on. The book is addressed to the broad sections of readers – students, post-graduates and scientific researchers – who deal with partial differential equations and optimal control. The work on book teaches to be modest, as far as the author realizes, in addition, how much he depends on others people. I would to thank my colleagues from the Institute of Cybernetics of National Academy of Science of Ukraine (NANU) and from the Kiev National Taras Shevchenko University, and also others people, who stimulated (knowingly or unknowingly) my work during long time. The first place in the list of my personal thanks belong to NANU academician Yu.M.Ermoliev and professor V.P.Didenko, who are responsible for the awakening of my interest to the theory of optimization and to the issues of the solvability of differential equations with non-smooth data. Also, I am very grateful to my colleagues, who has an influence on the content of my book: NANU academicians I.V.Sergienko, B.N.Pshenichny, N.Z.Shor, correspondent-members of NANU B.N.Bublik, V.S.Mel'nik, Yu.I.Samoilenko, V.V.Skopetzky, A.A.Chikriy, professors A.G.Burkovsky, Yu.M.Danilin, A.I.Egorov, N.F.Kirichenko, A.G.Nakonechny. I would to mark with gratitude the activity and enthusiasm shown by my young disciples, especially D.A.Nomirovsky and V.V.Semenov, whose results I used in the book in them kind consent. D.A.Nomirovsky and D.A.Klyushin translated the manuscript from Russian to English. The general editing of the book was performed by

xv D.A.Klyushin. Without him insistence (up to the willingness to share with author the responsibility for possible mistakes) the book would be never completed. Finally, I am very thankful to the employees, post-graduates and students of the Department of Computational Mathematics of the Faculty of Cybernetics and the Department of Differential and Integral Equations of the Mechanics and Mathematics Faculty of the Kiev National Taras Shevchenko University, and also to all listener of my lectures and the readers of the preliminary versions of this book. I am very grateful to the editors of the Kluwer Academic Publishers for their remarkable work. Especially, I would like to thank Ms. Angela Quilici for her kind help during the preparation of the book. Please, inform the author about all found mistakes and misprints via e-mail addresses: [email protected], [email protected].

S.I.Lyashko

Chapter 1 OPTIMIZATION OF LINEAR SYSTEMS WITH GENERALIZED CONTROL

1. FORMULATION OF OPTIMIZATION PROBLEM FOR DISTRIBUTED SYSTEMS AND AUXILIARY PROBLEMS Consider a system which functioning is described by linear partial differential equation [11, 14-17,19, 30, 65-73] in a tube domain

where

function depended on a spatial variable is a bounded domain in

is an unknown

and a time variable with smooth boundary

D(L) is a set of the functions which are sufficiently smooth in satisfy some conditions (bd) on the boundary of the domain Q . Let us consider that operator L mapping where

and into

is the space of functions, which are measurable

and square integrable on the set Q. Note that the set D(L) is dense in

operator

therefore we can correctly define an adjoint by Lagrange

with the domain of definition

which is the set of functions, which are sufficiently smooth in

and

satisfy adjoint boundary conditions Let the following chains of equipped Hilbert spaces are constructed with respect to

2

Chapter 1

where

are the completions of D(L) with respect to

some positive norms;

are the completions of

with respect to the same norms. Negative spaces

are

constructed with respect to and to the corresponding positive spaces. Hereinafter we shall obtain a priori estimations with respect to negative norms for various specific types of operators L :

where

is

the

operator adjoint to L by Lagrange, c, C here and below with indices and without them denote positive constants not depending on the functions u(t,x), v(t,x),

in addition, these embeddings are dense and operators of embedding are completely continuous. Spaces are Hilbert ones endowed with the norms satisfying the inequalities Denote by the inner product in and

bilinear forms constructed by extending on continuity to bilinear forms on respectively. By

we

understand analogous bilinear form. It follows from the above mentioned inequalities (2) that operator L ( respectively) may be extended on continuity to an operator

OPTIMIZATION OF LINEAR SYSTEMS ...

continuously mapping whole space space

(

(

3

respectively) into

respectively). It should been noted that the

inequalities (2) hold true for extended operators also (we shall save for them the previous denotations) for any For operators L and

the following identity holds true

The solutions of the original and the adjoint problems

we shall mean in the sense of the following definitions. D e f i n i t i o n 1. The solution of the problem (3) with a righthand side is a function for which there exists a sequence of such functions

that

D e f i n i t i o n 2 (strong solution). The strong solution of the problem (3) with a right-hand side is such a function that

in the space D e f i n i t i o n 3 (weak solution). The weak solution of the problem (3) with a right-hand side is such a function that the following equality

holds true for any functions

4

Chapter 1

In a similar way we introduce the definitions for solutions of the equation (4). L e m m a 1. Let the inequalities (2) hold true for the operators L and Then Definitions 1,2,3 are equivalent. P r o o f . We shall carry out the proof following the scheme Let u(t,x) be a solution of the problem (3) in the sense of Definition 1. Then taking into consideration the inequalities (2) with the help of passing to the limit we conclude that the value of the extended operator on the element u(t,x) obviously equals to F , i.e. Lu = F in the sense of the equality of the elements in the space Vice versa, let u(t,x) be a solution of the problem (3) in the sense of Definition 2. Choose an arbitrary sequence such that

As far as

Further, we have

the second term at the right-hand side

equals to 0. Taking into account (2) and the linearity of the operator L we obtain Thus, the equivalence of the Definitions 1 and 2 is proved. It is not difficult to prove also the equivalence of Definitions 2 and 3. Indeed, the statement is obvious. Let us prove that Let u(t,x) a the solution of the problem (3) in the sense of Definition 3:


By virtue of the arbitrariness of

5

we obtain that Lu = F

in

Remark. Analogous statements hold true for solutions of the adjoint equation (4). T h e o r e m 1. Let the inequalities (2) hold true for the operators of the problems (3) and (4). Then for any element there exists a unique solution of the problem (3) in the sense of Definitions 1-3, where constructed with respect to

is a negative space

and

P r o o f . Consider the functional

on functions

By the Schwarz inequality and (2) we obtain

Hence the functional l(v) may be considered as a linear continuous one dependent on

According to

the Hahn-Banach theorem [76] extend the functional continuously onto whole space

linearly and

On the basis of the theorem

about common form of a linear continuous functional defined in [77] there exists a function for any

such that

Consider this functional on the elements Then

6

Chapter 1

Hence, arbitrariness of

By virtue of the we obtain that Lu – F = 0 in the space

The uniqueness of the solution follows from the left-hand side of the first inequality in (2) and from the embedding The theorem is proved. T h e o r e m 2. Let the inequalities (2) hold true. Then for any element there exists a unique solution of the problem (4) in the sense of the analogues of Definitions 1-3 for the problem (4). P r o o f of Theorem 2 is similar to the proof of Theorem 1. It is possible to consider the generalized solution from more wide class. D e f i n i t i o n 4. The solution of the problem (3) with a righthand side

is a function

exists a sequence of such functions

for which there that

D e f i n i t i o n 5 (weak solution). The weak solution of the problem (3) with a right-hand side is a function such that the equality

holds true for any functions The generalized solutions of the problem (4) are defined in a similar way. L e m m a 2. Assume that the inequalities (2) hold true for the operators L and Then Definitions 4 and 5 are equivalent.


7

P r o o f . Let u(t,x) be a solution of the problem (3) in the sense of Definition 5,

and

Choose a sequence Then, if

such that is a solution of the problem

in in the

sense of Definition 1 (which exists by virtue of Theorem 1) then according to Lemma 1 we have

Hence,

and therefore,

i.e. the sequence

is fundamental in

Moreover, taking into consideration (2) we obtain

Thus,

is a fundamental sequence in

exists Further, we have

such that

hence, there

8

Chapter 1

Passing in the last equality to the limit as

we obtain

Taking into consideration the fact that the last equality holds true for arbitrary

and the relations (5), we state that in

also

and as far as

then

Granting (6), we convince oneself that

u(t,x) is the solution of the problem (3) in the sense of the Definition 4. Let us prove the inverse statement. Let u(t,x) is a solution of the problem (3) in the sense of Definition 4. Then

Taking into account (2), let us estimate the first and the third terms in the right-hand side of (7).

Passing to the limit in (7) as we obtain the required equality (5). T h e o r e m 3. Let the inequalities (2) hold true for the operators of the problems (3), (4). Then for any element there exists a unique solution of the problem (3) in the sense of Definitions 4-5.


P r o o f . The set

is dense in

there exists a sequence

9

Hence, for any element such that

It follows from Theorem 1 that for any function

there

exists a unique solution of the problem (3) in the sense of Definition 1. Using the inequality (2), we obtain

Thus, by virtue of the completeness of the space a function

there exists

On the other hand,

by Lemma 1 for any function

the following equalities hold

true or

Passing to the limit as

in the last equality, we obtain

Let us prove the uniqueness. Let be one more generalized solution of the problem (3) in the sense of Definitions 4-5. Then Theorem 1 implies that

So

in

The proof is complete.

10

Chapter 1

T h e o r e m 4. Let the inequalities (2) hold true for the operators of the problems (3), (4). Then for any element there exists a unique solution of the problem (4) in the sense of the analogies of Definitions 4-5 for the problem (4). P r o o f of Theorem 4 is similar to the proof of Theorem 3. L e m m a 3. Let u(t,x) be a solution of the equation (3) with a right-hand side

in the sense of Definitions 4-5, then

the following estimation holds true P r o o f . At first, we shall show that the following positive estimations follow from the a priori estimations with respect to the negative norms (2):

Indeed, by virtue of Theorem 3 there exists a unique generalized solution u(t,x) of the problem (3) with the right-hand side and this solution belongs to the space that for any

and for any

following equality holds true Using the Schwarz inequality we obtain

It follows from Lemma 2 such that

the


11

The Banach-Steinhaus theorem [76, 78] implies that the set of functions space

is bounded with respect to the norm of the and this proves that the inequality (10) holds true. The

inequality (9) is proved in a similar way. Let us prove that (8) holds true. Applying to the right-hand side of the equality (11) the Schwarz inequality and (10), we obtain

that implies (8). Remark. In a similar way we can prove the inequality where v(t, x) is a solution of the equation (4) with a right-hand side in the sense of Definition 4. L e m m a 4 . Let u(t,x) be a generalized solution of the problem (3) with a right-hand side F in the sense of Definitions 4-5 and Then u(t,x) is a generalized solution of this problem in the sense of Definitions 1-3. P r o o f . Let v(t,x) be an arbitrary smooth in function, which satisfies the boundary conditions Consider . We have

12

Chapter 1

where

is a sequence, which determines the solution u(t, x) by

Definition 4. Let us estimate each terms in the right-hand side. As far as v(t, x) is a smooth function, which satisfies the

conditions

then

Moreover,

i.e.

where v(t, x) is an arbitrary function from a set which is dense in Hence, Thus, u(t,x) is a generalized solution of the problem (3) by Definition 2. Remark. The similar statement holds true for solutions of the problem (4). Taking into account the above-mentioned facts, we shall further consider the following optimization problem. Let the functioning of a system is described by a partial linear differential equation: where

is a given element. The control of the system (12) is

carried out by choosing the controls h which are defined on a set of admissible controls from a reflexive Banach space of controls H;

is an operator, possibly non-linear,

On

the solutions of the equation (1) we define a functional


13

which is weakly lower semicontinuous with respect to a state of the system and which must be minimized on the set

2. EXISTENCE OF OPTIMAL CONTROL Consider the optimal control problem described in Section 1. Let a state of the system is determined by the equation (1.12): The state function u(t,x) depends on (via the right-hand side of the equation) from a control h defined on the set of admissible controls which belongs to the space of controls H . On the solutions of the equation we define some weakly semicontinuous with respect to the system state functional which must be minimized on the set Of course, it is necessary to require that the functional should been defined correctly of the solutions of the equation. For example, if

then the solution u(t,x,h) must belong to the space

then u(t,x,h) must belong to the Sobolev space

If

(such

situation we denote as and respectively). It is clear that the smoothness of the solution u(t,x,h) depends on the smoothness of the right-hand side of the equation (see Section 1). Therefore, the more wide space of mappings we consider, the more narrow class of functionals we may investigate.

14

Chapter 1

T h e o r e m 1. Let a system state be determined as a solution of the problem (1.12) under the following assumptions: 1) the performance criterion is a functional which is weakly lower semicontinuous with respect to the system state and below bounded; 2) the set of admissible controls is bounded, closed and convex in H ; 3) H is a reflexive Banach space; 4) is a weakly continuous operator mapping H into

5) the estimations (1.2) for the operators L and are valid. Then there exists the optimal control of system (1.12), i.e. such control

that

P r o o f . Choose a minimizing sequence of controls

i.e.

Since the set is weakly compact ( is bounded, closed and convex in the reflexive space H), we may extract from this sequence a subsequence which weakly convergences in H to some

As far as operator

the corresponding sequence

is weakly continuous,

is weakly convergent, and by

the Eberlein-Shmulyan theorem [78] this sequence is bounded in the space


15

Lemma 1.3 and above-mentioned reasoning imply that the sequence which corresponds to the subsequence is bounded in the norm of the space

and hence, we may extract

from it some weakly convergent subsequence

Granting both the facts that equation (1.12) when

for

in

is a solution of the

and the relations (1.5), we obtain

any function As far as

such that

where

weakly in H, and

continuous mapping of H into

is weakly

then

In addition, due to the fact that

Thus, passing to the limit in (2) as

we obtain

we conclude that

is a solution of the equation in the sense of Definition 1.5, and hence, in the sense of Definition 1.4 also. Since the performance criterion J(h) is weakly lower semicontinuous with respect to u(t, x) ,

16

Chapter 1

Hence, is the optimal control and Theorem 1 is proved. T h e o r e m 2. Let a system state be determined as a solution of the problem (1.12) under the following assumptions: 1) the performance criterion is a functional which is weakly lower semicontinuous with respect to the system state u(t,x,h) and below bounded; 2) the set of admissible controls is bounded, closed and convex in H ; 3) H is a reflexive Banach space; 4) is a weakly continuous operator mapping H into

5) the estimations (1.2) for the operators L and are valid. Then there exists the optimal control of system (1.12). P r o o f is similar to the proof of the previous theorem. Remark. As far as the operator is non-linear then the functional may be not convex and the optimal control may be non-unique. Consider the application of these theorems in the case of the optimization of distributed systems with point controls. The urgency of such investigations is stipulated both by the development of new technologies and by the simplicity of the control realization. In view of specific character of control in these problems it is possible to obtain more interesting results [80-86] Let the studied system be described by the linear partial differential equation Consider the optimal control problems for the systems, which are described by the equation (3) with right-hand sides in the following forms:


where is the Dirac function. Consider every variant of the mapping

where The control is where

17

separately.

and

is bounded, closed and convex set in

18

Chapter 1

By

we shall mean a functional defined

on smooth functions in

in the following way:

Suppose that

where is a completion of the space of smooth in functions with respect to the norm

and also let We shall prove that in this case extend

i.e. it is possible to

up to a linear continuous functional on

Indeed,

the linearity of the functional

is clear. Let us prove the boundedness of this functional. By the integral Cauchy inequality we obtain


Since the set

19

is bounded in

It is easily seen [74] that the following inequality holds true:

Granting this and (9), rewrite (11) in the following form:

Returning to the inequality (10), we finally obtain

that proves the boundedness of the functional smooth in

on the set of

functions v(t, x). Applying the Hahn-Banach theorem

we expand on continuity the functional

up to the linear

continuous functional which is defined on the whole space

20

Chapter 1

Thus, it is proved that

and hence, by

Theorem 1.3 there exists the unique generalized solution of the problem in the sense of Definition 1.4. Let us prove that the mapping weakly continuous.

sequence convergences are

is

be some weakly convergent in

As far as the weak and strong equivalent in the finite-dimensional strongly in under arbitrary

space, Put

Let us prove that for every

Indeed,

Since the set of smooth in

functions v(t, x) from

is dense in

we may take supremum only on these functions. Therefore, the last expression may be rewritten as


21

Let us estimate the numerator

Applying the integral Cauchy inequality to the right-hand side of this identity, we obtain

Taking into account (13), let us apply the Cauchy inequality once more:

By virtue of the inequalities relations (12) and (14), we obtain:

and the

22

Chapter 1

As far as

weakly converges to

in

by Eberlein-

Shmulyan theorem [78] the sequence of the norms

is

bounded, and hence

So,

Now, let us prove that for an arbitrary smooth in v(t, x) from

we have

Indeed,

Passing to the limit in the last equality is correct as far as weakly in

function


Due

to

the

fact

that

the

sequence

23

of the

norms

is bounded (proof is similar to the proof of the boundedness of

it is easily to prove that for an arbitrary

function

Taking into account all above-mentioned facts, we have

where Let inequality

The first term tends to zero, as far as by the Schwarz

The second term tends to function

i.e. the mapping

is weakly continuous.

Thus, for any

24

Chapter 1

Thus, we obtain T h e o r e m 3. Let a system state be determined as a solution of the problem (1.12) under the following assumptions: 1) the performance criterion is a functional which is weakly lower semicontinuous with respect to the system state u(t,x,h) and below bounded; 2) the set of admissible controls is bounded, closed and convex in H ; 3) 4) 5) the estimations (1.2) and (9) are valid. Then there exists the optimal control of the system (1.12). Remark. As it was shown above, there exists a sequence of controls which weakly converges to the optimal one Hence, the problem of optimization may be ill posed [88-92] in the above selected metrics, that can yields significant difficulties in computation of the optimal pulse control. This circumstance may be overcame by using the regularization of the control. Let be a convex, closed and bounded set in In this case from the sequence subsequence

we can extract a

which is weakly converges to

in

Due to the fact that is bounded and has sufficiently smooth bound, it is follows from Kondrashow theorem [27, 93] that the imbedding operator of into is completely continuous and,

hence,

the

sequence

is strongly


convergent in

to

25

Thus, the considered problem of

optimization will be well posed in In the similar way we can prove the following T h e o r e m 4. Let a system state be determined as a solution of the problem (1.12) under the following assumptions: is a functional 1) the performance criterion which is weakly lower semicontinuous with respect to the system state u(t,x,h) and below bounded; is bounded, closed and 2) the set of admissible controls convex in H ; 3) 4) 5) the estimations (1.2) are valid, and the following imbeddings

and inequalities hold true

Then there exists the optimal control of system (1.12).

2. where

is the k -th Sobolev derivative of the and The control is

where

is bounded, closed and convex in

26

Chapter 1

By

we

functional, which is defined on smooth in way:

shall

denote

a

functions in the following

where is the k -th derivative of the function v(t,x) with respect to the variable t . Suppose that

and pa

where is the completion of the space of smooth in functions with respect to the norm

and let

Similarly

to

the

previous cases i.e. the mapping

linear continuous functional on

and

we

can

prove that defines a

is weakly continuous

mapping, and hence, T h e o r e m 5 . Let a system state be determined as a solution of problem (1.12) under the following assumptions: 1) the performance criterion is a functional which is weakly lower semicontinuous with respect to the system state u(t,x,h) and below bounded;


2) the set of admissible controls and convex in H; 3)

27

is bounded, closed

4) 5) the estimations (1.2) and (15) are valid. Then there exists the optimal control of the system (1.12). In a similar way we can prove the following T h e o r e m 6 . Let a system state be defined as a solution of the problem (1.12) under the following assumptions: is a functional 1) the performance criterion which is weakly lower semicontinuous with respect to the system state u(t,x,h) and below bounded; 2) the set of admissible controls is bounded, closed and convex in H ; 3) 4) 5) the estimations (1.2) hold true, and also the following imbedding and inequalities are valid:

and Then there exists the optimal control of system (1.12). In studying of the optimal control existence in the case when the right-hand side of the equation is defined by one of the mappings we shall suppose for simplicity that the set with respect to the variable i.e.

is a tube domain

28

Chapter 1

3.

when The control is

where

is a bounded, closed and convex set in

By

we shall denote a

functional, which is defined on smooth in way:

Suppose that

functions in the following

and


and let

Similarly mapping

to

the

previous cases we can prove that Let us prove the weak continuity of the


Let

We shall prove that

Let v(t, x) is an arbitrary smooth in conditions

29

function, which satisfies the

Then similarly to the previous cases we obtain

Taking into account the boundedness of the set of functions and the definition of the negative norm, we obtain (17). Further proof is similar to the case of T h e o r e m 7. Let a system state be determine as a solution of the problem (1.12) under the following assumptions: 1) the performance criterion is a functional which is weakly lower semicontinuous with respect to the system state u(t,x,h) and below bounded; 2) the set of admissible controls is bounded, closed and convex in H ;

3) 4) 5) the estimations (1.2) and (16) are valid.

30

Chapter 1

Then there exists the optimal control of system (1.12). In a similar way we can prove the following T h e o r e m 8. Let a system state be determined as a solution of the problem (1.12) under the following assumptions: 1) the performance criterion is a functional which is weakly lower semicontinuous with respect to the system state u(t,x,h) and below bounded; 2) the set of admissible controls is bounded, closed and convex in H ; 3) 4) 5) the estimations (1.2) hold true, and also the following imbeddings and inequalities are valid:

Then there exists the optimal control of the system (1.12). 4. where The control is

where

is a closed and bounded set in


31

By

we shall denote a functional defined on smooth in following way:

Suppose that

functions in the

and


and let

Completely similarly to the three previous cases it is possible to prove that

Let us prove that in

is a weakly continuous mapping. Let We shall prove that for any i, j

Similarly to the previous cases let us consider

32

Chapter 1

where v(t, x) is an arbitrary smooth in function. Using the Newton-Leibniz formula, we obtain

Applying the integral Cauchy inequality and the inequality , we have


33

The Sobolev imbedding theorems imply that for any function

the following inequality holds true [74]

Therefore,

and

Substituting (22) and (23) into (20), we obtain

Note, that as far as the sequence hence, it is bounded, we have

is weakly convergent, and

34

Chapter 1

Whence,

Finally, we obtain

Now, let v(t, x) is an arbitrary function from the following identity holds true

It is clear that


35

It follows from the previous reasoning that the first term tends to zero. Since weakly converges to zero in the second term tends to zero also. Thus, we have proved that the mapping is weakly continuous. T h e o r e m 9. Let a system state be determined as a solution of the problem (1.12) under the following assumptions: is a functional 1) the performance criterion , which is weakly lower semicontinuous with respect to the system state u(t,x,h) and below bounded; is bounded, closed and 2) the set of admissible controls convex in H ;

3) 4) 5) the estimations (1.2) and (18) are valid. Then there exists the optimal control of the system (1.12). In a similar way we can prove the following T h e o r e m 1 0 . Let a system state be determined as a solution of the problem (1.12) under the following assumptions: is a functional 1) the performance criterion which is weakly lower semicontinuous with respect to the system state u(t,x,h) and below bounded; is bounded, closed 2) the set of admissible controls and convex in H ; 3) 4)

36

Chapter 1

5) the estimations (1.2) are valid, and also the following imbeddings and inequalities hold true

Then there exists the optimal control of the system (1.12). 5.

The control is

By

we shall denote a functional defined on smooth in

functions in the following way:

where v(t, x) is a smooth in

function.

Suppose that

and


and let We shall prove that the functional the integral Cauchy inequality, we obtain

is bounded. Indeed, using


37

Using inequalities (21) and (24), we have

that proves the boundedness of the functional

the linearity of

is obvious. Let us prove that

where

is some weakly convergent in

sequence:

that being weakly convergent the sequence the space into

is bounded in

and since the imbedding of the space is compact, we obtain

with respect to the norm of the space

The sequence of the functions converges to

Note,

in

and, hence,

weakly

38

Chapter 1

At first, let us prove that

To do this let us consider

for any smooth in formula we obtain

function v(t, x). Applying the Newton-Leibniz

Applying the integral Cauchy inequality to the right-hand side, we obtain:


39

Thus, we have

As far as

is weakly convergent, then the

sequence of the norms set of smooth in

is bounded. Granting that the

functions v(t, x) is dense in

The last expression implies that

To finish the proof we have to verify that for any

we have

40

Chapter 1

Indeed,

as far as

weakly converges to

in

Thus, taking into account (25) and (26), we obtain

for an arbitrary function In other words, the mapping

is weakly continuous. T h e o r e m 1 1 . Let a system state be determined as a solution of the problem (1.12) under the following assumptions: 1) the performance criterion is a functional which is weakly lower semicontinuous with respect to the system state u(t,x,h) and below bounded;


2) the set of admissible controls and convex in H ;

41

is bounded, closed

3) 4) 5) the estimations (1.2) and (24) are valid. Then there exists the optimal control of the system (1.12). Similarly we can prove the following T h e o r e m 1 2 . Let a system state be determined as a solution of the problem (1.12) under the following assumptions: is a functional 1) the performance criterion which is weakly lower semicontinuous with respect to the system state u(t, x, h) and below bounded; is bounded, closed 2) the set of admissible controls and convex in H ; 3) 4) 5) the estimations (1.2) are valid, and also the following imbeddings and inequalities are valid:

Then there exists the optimal control of the system (1.12). Note, that we may consider the problem of optimal control when the right-hand side of the state equation is a linear combination of the functionals Consider, for example, the problem

42

Chapter 1

where The control is

By

we shall denote a functional defined on smooth in following way:

Suppose that

functions in the

and

and let As far as

are

negative spaces constructed on the pairs similarly to the cases

and and

we shall

prove that the right-hand side of (27) belongs to the space and hence, to the Let us prove that mapping. Indeed, let

is a weakly continuous is a weakly convergent in

In is clear that

sequence


43

where

in

That is why from the results of this section we have that in in Since

the bilinear forms coincide for the pairs from

and hence,

That is why

weakly in

and hence, in

also. Thus, the following

theorem holds true: T h e o r e m 1 3 . Let a system state be determined as a solution of the problem (1.12) under the following assumptions: is a functional 1) the performance criterion which is weakly lower semicontinuous with respect to the system state u(t, x, h) and below bounded; is bounded, closed 2) the set of admissible controls and convex in H ;

3) 4) 5) the estimations (1.2) and (28) are valid. Then there exists the optimal control of system (1.12). In a similar way we can prove the following

44

Chapter 1

T h e o r e m 1 4 . Let a system state be determined as a solution of the problem (1.12) under the following assumptions: is a functional 1) the performance criterion which is weakly lower semicontinuous with respect to the system state u(t, x, h) and below bounded; is bounded, closed 2) the set of admissible controls and convex in H ; 3) 4) 5) the estimations (1.2) hold true, and also the following imbeddings and inequalities are valid

Then, there exist optimal controls of the system (1.12).

Chapter 2 GENERAL PRINCIPLES OF INVESTIGATION OF LINEAR SYSTEMS WITH GENERALIZED CONTROL

1. DIFFERENTIAL PROPERTIES OF PERFORMANCE CRITERION Depending on the properties of the operators L and it is possible to consider various controls and performance criteria. Hereinafter we shall study the differential properties of integral, quadratic with respect to the system state performance criterion for the problem of pulse optimal control and later we shall generalized the corresponding theorems for the case of the problem of the optimal control with an arbitrary right-hand side. In the case when the following inequalities are valid for the operator L

and the right-hand side

of the equation (1.1.12) is of the form (the

first

case

considered in Section 1.1.2) we consider the performance criterion:

where

is a function describing the desired functioning of the

system, The functional (1) is defined correctly as far as Theîrem 1.1.3 guarantees that the function u(t,x,h) belongs to the space

46

Chapter 2

Suppose that

and moreover this

imbedding is continuous. T h e o r e m 1. Provides that conditions above mentioned are satisfied, the functional (1) is differentiable by Gâteaux in the space and its gradient is of the form:

where

is a control,

v(t, x) is a solution of the problem

Proof.

Let

and be some elements belonging to

and are solutions of the boundary problem (1.1.12) corresponding to these controls. Denote Then the increment of the performance criterion may be represented in the following form

Define the adjoint state as a solution of the problem

GENERAL PRINCIPLES OF INVESTIGATION ...

47

It follows from the results of Chapter 1 that there exists a unique solution of this problem in the class of functions belonging to Then (2) we may rewrite in the following form Obviously,

is a solution of the problem

As a consequence of the results obtained in Chapter 1 there exists a solution of this problem which is defined as a function such that for all

(including v(t, x)) the

following equality is true:

We obtain

Represent the increments

We obtain

from (3) as

48

Chapter 2

We shall prove that

this fact proves the theorem. Here

is an inner

product in H . To do this we shall show that inequality

Using the

we obtain a chain of the following

inequalities

According to the definition of the negative norm in continuity of the imbedding

we obtain

and


49

It follows from this that The equality (5) implies that investigated performance criterion is differentiable by Gâteaux in and its partial derivatives are determined by the following expressions

Thus, the theorem is proved. T h e o r e m 2. The performance continuously differentiable in Proof.

criterion

is

50

Chapter 2

We shall prove that when

the

expression in the right-hand side (8) also tends to zero. Let us convert this expression adding and subtracting from each term in the first square brackets the following expression

and from each term of the second sum —

Then, using the obvious inequalities

where (a,b,c) are real numbers, we obtain


Denote the sums in the right-hand side of (9) by required to prove that

It is

as

L e m m a 1. Let

Then for any

51


the following inequality holds true

P r o o f of Lemma 1 for specific operator follows from the positive inequalities, which are valid as a result of the inequalities (1.1.2). Denote Then


where Obviously that


Using Lemma 1 and the inequality which validity were proved in Lemma 1.1.3, we obtain a chain of the following inequalities

52

Chapter 2

According to the definition of the norm in of the set of the admissible controls

and to boundedness

we obtain

Here we have used the integral Cauchy inequality. Similarly, estimating the i-th term of the second summand in the right-hand side (10), we obtain Summing (11) and (12) with respect to i from 1 to N and substituting into (10), we obtain

where As far as the expression in the left-hand side of (13) is the i-th term in the first sum in the right-hand side of (9) then with The proof of the fact that

with

It follows from Lemma 1 and boundedness of

follows from that


for all

Then applying to

obtain The fact that

53

the Cauchy inequality, we

is estimated similarly to as

follows from the continuity

of the function v(t, x) with respect to the variable t. It follows from the proven theorem that the performance criterion J(h) is continuously differentiable by Gâteaux, and hence, by Fréchet and it is possible to express the necessary conditions in the following way

where is the required solution. In the similar way it is possible to investigate others problems of the singular optimal control. Let us generalize the results of the previous theorems for the case of the problem of the optimal control with an arbitrary right-hand side. Let a state of some system is described by the equation (1.1.12) and for the operator L the following inequalities hold true

Suppose that the performance criterion is of the following form

54

Chapter 2

where

are some known functions from

respectively, and T h e o r e m 3. Let the state of the system is determined as a solution of the equation (1.12) with the righthand side The performance criterion is of the form (14). Then, if there exists the Fréchet derivative of the mapping

in some point

then the

performance functional J(h) is also differentiable by Fréchet in the point

, and the derivative is determined by the expression

where v(t, x) is a solution of the adjoint problem

P r o o f . It is clear that the expression far as the equation (16) has a unique solution (Theorem 1.1.2) and

for any Let

is an increment of a control. Consider

is correct as


Let

It is obvious that and hence, as far as

Moreover, using (17)

Therefore

Then

is a solution of the following problem: then

we obtain

55

56

Chapter 2

As far as

the Fréchet derivative of the

mapping inequality

then

: it follows from the

that

On the other hand, using the inequality

of

Lemma 1.1.3 we obtain

Applying the inequality

and (18), we obtain


57

Let

If

then the inequality (18) holds true and Thus, finally we obtain

as required.

T h e o r e m 4. Let the mapping

has a

Fréchet derivative which is continuous in the point Then the derivative is continuous in the point also. P r o o f . Choose an arbitrary number and let h is an arbitrary point from

which is close enough to

in order that

derivative exist, and hence, by Theorem 3 the derivative exists also. Consider

58

Chapter 2

where sides Since

and

are solutions of the problem (16) with the right-hand and

is continuous at the point

sufficiently small neighbourhood of the point namely from this neighbourhood) in which bounded

respectively. , there exists

(h we shall choose is

. Let also

Applying the Schwarz inequality we obtain

Note that it follows from Theorem 1.1.2 and from the proof of Lemma 1.1.3 that , where v(t, x) is a solution of the equation (1.1.4). Therefore


where

59

(constant M exists by virtue of continuity,

and hence by virtue of boundedness of the functions Using the inequality

from Lemma 1.1.3 we have

Note that since then

are some known functions from and hence

Let us apply to the term finite increments:

in

the formula of the

60

Chapter 2

As a result we get

Whence choosing

we have

The theorem is proved. T h e o r e m 5. Let a mapping

has a Fréchet

derivative in some bounded convex neighbourhood of a point which satisfies the Lipschitz condition with an index from the neighbourhood of the point that

such

Then the derivative

satisfies the Lipschitz condition with the index P r o o f . Let us prove that it follows from the Theorem assumptions that the derivative is bounded in the neighbourhood of the point

Indeed, let h is an arbitrary point from the

neighbourhood of the point inequalities hold true

Then, obviously, the following

where d is the diameter of the neighbourhood of the point

GENERAL PRINCIPLES OF INVESTIGATION ... Now, let Consider

where

and

61

be arbitrary points from this neighbourhood.

are solutions of the adjoint problems (16) with the

right-hand sides

and

,

respectively. Applying the Schwarz inequality we get

Note that it follows from the analogy of Theorem 1.1.1 and from the proof of Lemma 1.1.3 that , where v(t, x) is a solution of the equation (1.1.4). Therefore

62

Chapter 2

where

in

Using

the

inequality

from Lemma 1.1.3 we obtain

Note that as far as then

are some known functions from . Moreover,

and applying the formula of finite increments to the first term we get

taking into consideration that i.e.

, we get

Hence,

belong to the bounded set


Let us apply to the term

63

the formula of finite

increments:

As a result we obtain

Whence,

The theorem is proved. Remark. Granting that the mapping form

and the functional

is of the does not depend

on the control we may rewrite the formula (15), which define the gradient, in the following form

where

is a Fréchet derivative of the mapping , the function v(t, x) is chosen in a similar way as

in Theorem 3. Respectively, in Theorems 3 and 4 we may require

64

Chapter 2

that the assumptions of this theorem hold true for the mapping instead of Consider the application of the theorems of this section for the cases of the right-hand sides of specific types. We shall consider functions defined in Chapter 1. 1. Let and where


is a convex, closed and bounded set in

Suppose that

and


and also, let Make sure that the mapping

is differentiable by Fréchet and its partial derivatives are of the form

Indeed, it is clear, that the expression


65

defines a linear continuous operator Let us estimate

Consider the numerator of this ratio. At first, let v(t, x) be some smooth in

function satisfying the boundary conditions

Let us use the Taylor formula.

where Granting that the following inequality holds true

Then

66

Chapter 2

we get

Applying the integral Cauchy inequality we obtain:

Using the fact that the set of the considered smooth functions v(t, x) is dense in we have

It is clear that implies

such that the inequality


67

Hence, the mapping has a Fréchet derivative and by Theorem 3 the functional J(h) in the problem of the optimal control (1.1.12) with the right-hand side also. Let us prove that

has a Fréchet derivative

satisfies the Lipschitz condition with the

index Indeed,

Consider the numerator of this ratio. Let v(t, x) be a smooth in function satisfying the conditions

68

Chapter 2

Let us use the integral Cauchy inequality:

Again, we use the integral Cauchy inequality:


69

Granting that the set of the considered functions v(t, x) is dense in we get that

Whence,

as far as Thus,

, and

is bounded.

satisfies the Lipschitz condition with the index By Theorem 5 we may state that the derivative by Fréchet

satisfies the Lipschitz condition with the index hence

and

is continuous with respect to h . Note that we may repeat

the previous reasoning concerning to the mapping Theorem 3) and we can prove that

(including

70

Chapter 2

i.e.

with respect to the direction

satisfies the Lipschitz

condition (with the index 1). 2. Let

where

is the

k -th Sobolev derivative of the

and

The control is where


Suppose that

and


and let

Similarly to the previous facts we can prove that the mapping has a Fréchet derivative, which satisfy the Lipschitz condition with the index Applying Theorems 3 and 5 we state that the performance criterion

is differentiable by Fréchet, and the corresponding derivatives are defined by the expressions:


71


The derivative

satisfies the Lipschitz conditions with the

index and with respect to the direction - with the index 1. Consider the application of Theorems 3-5 of this paragraph in the case when the right-hand side of the equation (1.1.12) is defined as As in previous section we suppose that the set Q is a tube with respect to the variable and which is bounded, closed and convex set from the Hilbert space of controls H. 3. Let the right-hand side of the equation is of the form

where

The control

is

where


Suppose that

where

and

is a completion of the space of smooth in

functions with respect to the norm

72

Chapter 2

and let

Similarly to the case of

which was considered above it is

easy to see that the mapping

has a Fréchet

derivative, which is defined as:

We shall prove that index Consider

Let v(t, x) be a smooth in numerator of the ratio:

satisfies the Lipschitz condition with the

function from

. Consider the


Using the inequality

and the integral Cauchy inequality we get

Again use the Cauchy inequality:

73

74

Chapter 2

Whence we obtain

Taking

into

account

that

Therefore, with the index

is

bounded

we

have

satisfies the Lipschitz condition

with respect to h . By Theorems 3-5 the

performance functional has a strong derivative

in the region

which satisfy the Holder condition with the index

with

respect to h and is defined by the expression Considering directly the performance functional that

i.e. with respect to the direction Lipschitz condition with the index 1 .

the functional

we can prove

satisfies the


75

4. Let


where Under


we shall mean the following functional:

where v(t,x) is a smooth in Suppose that

function. and

where is a completion of the space of smooth in with respect to the norm:

functions

and also let Similarly to the previous facts it is easy to see that strong derivative:

has a

76

Chapter 2

and also that the derivative

satisfies the Hölder-Lipschitz

condition with the index and the performance functional has a Fréchet derivative which also satisfies the Hölder-Lipschitz condition. Consider the last case when the right-hand side of the equation is defined as 5. Let

The control is

Under

we shall mean the following functional

where v(t,x) is a smooth in Suppose that

function. and


where

77

is a completion of the space of smooth in

functions with respect to the norm

Make sure that the mapping

is strongly differentiable and the

derivative is defined by the expression

Indeed,

Whence

Using the fact that Sobolev space, we get

where . Therefore,

is the

78

Chapter 2

This fact proves that

is strongly differentiable. Similarly to

the previous facts we make sure that

satisfies the Lipschitz

condition with the index

2. REGULARIZATION OF CONTROL In the previous section the questions of performance, criterion gradient existence and its continuity depend upon differential properties of the right-hand side of state equation have been studied. In order to calculate the performance criterion gradient in any point, it is necessary to solve the direct and adjoint problems. In this paragraph we shall study the optimization problem in the case when optimal control exists, but the right-hand side are not differentiable enough for calculation of performance criterion gradient [80,94]. This difficulty can be remedied by regularization of control. As in the previous sections we assume here that the operators L and satisfy the inequality in the negative norms


79

Consider the following averaging of the distribution F in the righthand side:

where

By the integral

we mean the integral in the

sense of the distribution theory. Instead of using distribution averaging, we may consider a sequence of functions when Consider a regularized problem of impulse optimal control.

where

We assume that

(although all the theorems can be

generalized to the case The vector optimization problem (l)-(3), where

is a control of the

80

Chapter 2 is bounded, convex and closed set of admissible

controls. T h e o r e m 1. There exists an optimal control (generally not unique) of the optimization problem (1)-(3). P r o o f . Let be the minimizing sequence of the problem (l)-(3). This means that

Since the set

is bounded, convex, closed set and is weakly compact in

compact in

. Set

is Hilbert, is

Thus, there exists a subsequence (it is denoted by

again) such that

in

weakly in Let us introduce the following notation

We shall show that

weakly in For every function

By the Schwarz inequality

and each number

we have


Since

81

in

and

the first term of the right-hand side of (5) vanishes as The second term of the right-hand side of (5) vanishes also, because

weakly in

Thus the formula (4) has been proved. In accordance with the results of Section 1, a solution

(in

the sense of Definition 1.1.1) of the problem (1) with the right-hand side the equation

exists and unique. The solution

satisfies

for all

Taking into account that the set

is bounded, we have

It follows from the inequality (7) that there exists weakly convergent to . Passing to the limit as

for all

and so that in

subsequence

in the equality (6), we obtain

82

Chapter 2 If we take into consideration the above equality, we conclude that is solution of the problem (1) with the right-hand side

Since square of norm in the Hilbert space is weakly lower semicontinuous, the functional (2) is weakly lower semicontinuous in

domain

Consequently,

is an optimal control of the problem (l)-(3).

T h e o r e m 2. The set of the optimal controls problems (1)-(3) contains a weakly convergent The sequence

of the sequence

weakly converges to an

optimal control

of the initial (non-regularized) problem,

the points of impulse

being strongly convergent to

P r o o f . We shall show that at the fixed control

where respectively. For all

we have

– solutions of the regularized and initial problems, we have

Taking into account the definition of delta function we obtain


83

Summing up the above inequalities over and considering inequality of Lemma 1.1.3, we have (8). By Theorem 1 for all there exists an optimal control of the regularized problem. By virtue of the boundedness of the set

the set

subsequence

is weakly compact, hence, there exists a which weakly converges to

statement

The

becomes apparent when the

equivalence between convergence in norm and weak convergence in the finite-dimensional space is considered. From the weak convergence

of

we weakly in

Reasoning similarly, we have

weakly in

find,

as

well,

that

84

Chapter 2

Adding up these three relations, we have that weakly in Thus, we prove that sequence weakly

convergent

in

the

is , space

Therefore,

is bounded. By applying the inequality of Lemma 1.1.3 we conclude that the sequence problem (l)-(3) with right-hand sides

of solutions of the

is also bounded, so that the sequence

contains a weakly

convergent subsequence (which is denoted by again). Since weakly in it is easy to prove that weak limit of the subsequence

in the space

is the solution

of the initial optimization problem with the right-hand side From this it follows that

The statement

becomes apparent when it is

considered that for all space

problem.

Let

When

in the norm of the be an optimal control of the initial optimization

it

is

taken

into

account

the following relation is valid

that


85

From (9) and (10) we obtain

Therefore,

and

the initial optimization problem as well as may be not unique, the controls and Remark. If instead of then the sequence

is the optimal control of Since the optimal control may be different.

we consider

converges to the

in norm

T h e o r e m 3. There exists a Fréchet derivative of the performance criterion The Fréchet derivative is of the following form grad

where

is a solution of the adjoint problem with the right-

hand side P r o o f . We denote by an increment of a control h . Corresponding solution of the system (1) has an increment An increment of the performance criterion

can be represented in the form

86

Chapter 2

Adjoint state is defined as a solution of the following problem As appears from Section 1.1, there exists a unique solution of this problem belonging to the space

The increment

can be

transformed to the form

It is obvious that the increment following problem

is a solution of the

The results of Section 1 ensure the existence and uniqueness of the solution of the above problem. The solution

satisfies the

following equation

for all

Taking into account this relation, we have

An increment the form

can be represented in


87

Considering this equation, we have

Taking into account the estimation

we have

which is what had to be proved. The function can be approximated, very useful from practical point of view, by the sum

where

An analogous theorem for such way of approximation holds true. T h e o r e m 4. There exists a Fréchet derivative of the performance criterion of the optimization problem with the right-hand side This derivative satisfies the Lipschitz condition and can be represented in the following form

88

Chapter 2

where

is a solution of the adjoint problem with the right-

hand side P r o o f . In the same manner theorem we denote by an corresponding solution of the An criterion

as in the increment system (1) increment

can be represented in the form

As usual, we use the adjoint state and obtain

Divide (12) by

(12) and pass to the limit as

case of the previous of control h . The has an increment of the performance


89

which is what had to be proved in the first part of the theorem. The Lipschitz condition of the performance criterion gradient follows from the a priori inequalities in the negative norms and the inequality We shall first prove the inequality (14) for a smooth function In this case we have

Square the right and left-hand sides and integrate over the domain

which is what had to be proved in (14) for an arbitrary smooth function The inequality (14) for all

is proved by expansion by

continuity. Next, in the same manner as in Section 2.1 we generalize abovementioned results for the systems with the right-hand side of the general form. As in the previous case, we assume that the operators L, satisfies the inequalities in the negative norms

90

Chapter 2

We also assume that

and the functional

is defined at every function u(t, x) from Consider the regularized problem where It is required to minimize the functional T h e o r e m 5. Consider the optimal control problems (1.1.12) and (1)-(3). If 1) the admissible set of controls is convex, closed, and bounded in the Hilbert space H ; 2) maps satisfy the conditions:

for

a) an arbitrary sequence b)

is weakly continuous,

c) 3) the performance criterion weakly lower semicontinuous; then there exist optimal controls

for all fixed is upper semicontinuous and of the problems (1.1.12)

and (1)-(3), there exists a weakly convergent subsequence and an arbitrary weakly convergent subsequence

converges

to

weakly in H . P r o o f . According to Theorem 1.2.1, the conditions 1), 2a), and 3) ensure the existence of optimal controls of the problems (1.1.12)

and (l)-(3). Consider a set Since the set is bounded in the Hilbert space H, there exists a weakly convergent to


subsequence

Consider an arbitrary weakly

convergent subsequence and convex, the element solutions

Taking

91

Since the set belongs to

is closed

Consider the sequence of

By Lemma 1.1.3, we have

into

account the condition 2a), we have that weakly in Therefore, the sequence of

the norms

is bounded, so there exists a weakly

convergent subsequence

Since

are

solutions, we have

for all

Passing to the limit as

in (17), we have

It is now clear that the element u is a solution of the problem (1.1.12) with the right-hand side But there exists an unique solution of the problem (1.1.12); therefore entire sequence converges weakly to any other function

If there exists a weakly convergent to subsequence

indeed, then we

can prove in much the same way, that is an other solution of the problem (1.1.12) with the right-hand side contrary to the uniqueness of the solution. Since the functional is weakly lower semicontinuous, we have

92

Chapter 2

for all

Consider the sequence

Therefore, functional

By Lemma 1.1.3, we have

in the space

Since the

is upper semicontinuous, it follows that

Employing the relations (19) and (20), we have

for all

To put it in another way, the control is optimal. Consider the applications of this theorem. 1. Let there be given the right-hand side of the equation (1.1.12) in the following form:

where The control is where the admissible set Suppose that

We also assume that

is bounded, closed and convex in and


93

Set where

The regularized control is

It is not difficult to prove that

Let us verity that maps

satisfy the conditions 2a)-

2c) of Theorem 5, from whence it follows that maps satisfy the same conditions 2a)-2c). We shall test the condition 2a). Let

be an arbitrary weakly

convergent sequence. Then

Let v(t, x) be an arbitrary smooth function in the domain satisfies the conditions

We have

that

94

Chapter 2

By the mean value theorem, we obtain

where Consider the term

By the Schwarz inequality, we have:


95

Since,

The expression smooth in

vanishes for an arbitrary

function v(t, x). This set of functions is dense in

We shall show that Let v(t, x) be an arbitrary smooth in function. Considering the previous reasoning and the Schwarz inequality, we have


we obtain

96

Chapter 2

It follows from the weak convergence of

that the sequence

is bounded. Taking into account relation (22), we arrive at the following inequality.

Thus,

for all smooth in

functions v(t, x).

Considering that this set of functions v(t, x) is dense in

we

claim that

The weak convergence of a sequence of functionals follows from the pointwise convergence of a sequence of functionals in the dense set and boundedness of the sequence of norms of functional. We conclude that the condition 2a) of Theorem 5 is proved. Reasoning similarly, we convince that map is weakly convergent and

for h .

Thus, we have proved the following theorem T h e o r e m 6. Consider the optimal control problems (1.1.12) and (15) with the right-hand sides of (21), (23), respectively. If 1) the admissible set of control is convex, closed, and bounded in the Hilbert space H ;


97

2) the performance criterion is upper semicontinuous and weakly lower semicontinuous, then there exist optimal controls of the problems (1.1.12), (21) and (15), (23), there exists a weakly convergent subsequence and an arbitrary weakly convergent subsequence converges to weakly in H . Remark. The performance

satisfies the conditions of Theorem 6. Consider the problem of the performance criterion differentiability in the regularized optimal control problems. The existence of an optimal control of the regularized problems follows from the results of the previous section. We shall prove that there exists a Fréchet derivative of a map

To prove this formulas, consider

98

Chapter 2

If a value of is sufficiently small, the first and second integrals have intersecting intervals of integration:

Consider every summand of the right-hand side. By the mean value theorem, we have


99

where We can now easily show that an order of smallness of the first integral of the right-hand side equals to

By the Schwarz inequality,

whence

Taking the analogous transformation of the second integral of the right-hand side of (26), we have

100

Chapter 2

Finally, consider the third integral of (26)

We shall prove that the first and third integrals have the second infinitesimal order of control. By the mean value theorem and the inequality (24), we obtain

Reasoning similarly, it is easy to make sure that the third integral has the second infinitesimal order of control. Returning to the formula (26), we obtain


101

or

as required in (25). Consider the property of smoothness of a Fréchet derivative For this purpose we study the norm

Analyse the numerator

Rearrange the summand in the following way:

102

Since the set of smooth functions v(t, x) is dense in

Chapter 2

consider

only smooth function v(t, x). Consider all summands in the right-hand side separately:

By the Schwarz inequality and the inequality (24), we obtain


Next, we have

Analogously, consider the second summand of (29)

Examine the third summand of (29)


103

104

Chapter 2

Returning to (29), we obtain

so that the equality (28) can be rewritten in the form To put it in another way, the derivative Lipschitz condition with index

satisfies the

By Theorem 1.5, we assert that

gradient of regularized performance criterion satisfies Lipschitz condition with index

also. When performance criterion gradient

is analysed directly (without Theorem 1.5), it should be seen that the performance criterion directional gradient with correspondence to direction satisfies the Lipschitz condition with index Analogously, we could study the other right-hand sides of the equation (1.1.12). Let us show the specific cases of the regularization of the righthand side.


105

2. be the right-hand side of the equation, and

where

Assume that

and

3.

where

Assume that

4.

and

106

Chapter 2

where

Assume that

and

5.

where

Assume that

and

Chapter 3 NUMERICAL METHODS OF OPTIMIZATION OF LINEAR SYSTEMS WITH GENERALIZED CONTROL

1. PARAMETRIZATION OF CONTROL In this section applying procedure of parametrization, the minimization problem of the performance criterion J(h) in an infinitedimensional space we replace with the corresponding minimization problem in a finite-dimensional space. This substitution allows us to decrease the computational complexity of the implementation of the gradient methods. Let the system state be a solution of the following initial boundary value problem

where

is an orthonormalized basis in numbers

are such that

and

is

closed, convex, and bounded set in Let

and c be a

matrix with rows

We denote by the space of control. It is required to find a minimum of the following functional

108

Chapter 3

in the admissible set of control

where

is closed, convex and bounded set in T h e o r e m 1. There

exists

an

optimal where

control is

defined in (2). P r o o f. Let

be a minimizing sequence. Since

is

bounded, convex, closed set, the set ' is weakly compact. Thus there exists a weakly convergence subsequence (it is denoted by again). Since H is finite-dimensional space, the subsequence converges to number

strongly in H, whence for any

we have

Similarly to the previous sections we can prove that

in

The fact that a norm is a continuous functional in a Hilbert space implies that functional (2) reaches its infimum in the admissible set (the space H is finite-dimensional). Note that there exists a subsequence

such that

strongly in weakly in

where element

is a solution of

the initial optimization problem and T h e o r e m 2.

If

(imbedding is

continuous), then performance criterion in the space

is differentiable

and its gradient is of the following form

NUMERICAL METHODS OF OPTIMIZATION...

109

where v(t, x) is a solution of the adjoint problem with the righthand side Proof. elements

Let from

and the

admissible

are arbitrary set

and

functions

are corresponding states of the system (1). Denote an increment of the solution by Then an increment of the performance criterion (2) is of the following form

Define the adjoint state as a solution of the following equation It follows from Section 1 that there exists the unique solution of this equation. Therefore, the equation (3) can be rewritten in the form It is obvious that the increment

is a solution of the equation

110

Chapter 3

It follows from Section 1 that there exists the unique solution of this equation such that

for any

Let y = v. Then using the previous equation, we have

The increment following form

can be represented in the

Substituting this equality in (4), we obtain

To prove the theorem, it is sufficient to show that


where

111

is a inner product in H.

For this purpose we shall show that the inequality

By

we obtain

By the definition of the norm in the space the imbedding

and continuity of

we conclude

112

Chapter 3

This proves that

as required.

It follows from the inequality (5) that the performance criterion has a Gâteaux derivative in the following form

and its partial derivative is of

Remark 1. In the case of parametrization of other right-hand sides of the state equation the analogous theorems hold true.

2. PULSE OPTIMIZATION PROBLEM Earlier the pulse and point-pulse optimization problems were studied for some distributed systems. It was shown that a priori estimates in the negative norms allows us to prove the existence of the optimal controls of the investigated systems, to study the questions of its controllability, and to write out explicitly the gradient of the performance criterion of the original or regularized problems. In this section gradient methods are proposed for solving the problems of optimal control [95-100], which use an approximation of the performance criterion at every iteration [82, 101]. This approach is based on idea of solving the limit extremum problems [22, 102, 103]. At first, consider an optimization problem with pulse impact. Suppose that system state satisfies the following linear equation

NUMERICAL METHODS OF OPTIMIZATION.. .

113

where L maps In our case the right-hand size of the equation (1) is

where

is the Dirac delta-function. It is requested to find time

moments of pulse impacts

on an admissible set

which minimize the following performance criterion

where

is a known element from

is closed and

bounded set from Suppose that a priori estimates in the negative norm are valid:

where is formally adjoint operator. As it has been shown in the previous chapters, estimates (2) enable us to prove the existence and uniqueness theorem for the generalized solution of original and adjoint problems and the existence theorem for the optimal control h*. If in addition the following imbedding is valid, then we can find the performance criterion gradient explicitly. Otherwise, we regularize the original problem. There exist optimal controls of regularized problems and these controls converge to the optimal controls of the original optimization problem as the parameter of regularization vanishes. The regularized problem is of the following form

114

where

Chapter 3

when

By the integral (6) we mean the bilinear form in the sense of the theory of distributions. Hereinafter we shall omit in the regularized problem. In order to find the optimal control we may use different gradient methods [104, 105]. It follows from the equation for gradient that to find the gradient J'(h) it is necessary to solve direct and adjoint problems and to differentiate and integrate some expressions. As a rule, these procedures are implemented by numerical methods thus we have only a uniform convergent sequence of approximations instead of the exact value of J'(h). Uniform convergence follows from the a priori inequalities in the negative norms, where the constants do not depend on h. The inequalities (2) enable us to consider the minimization problem for some functional, which is equivalent to solving the initial boundary value problem [62, 64,106]. In the following sections we shall also discuss a projective numerical method for solving the initial boundary value problem. We shall build the procedures of minimization of gradient that use the approximation on every iteration.


115

Note that the similar results were obtained for the problem of nonlinear programming in [24, 25]. In these papers the convex functional J(h) was approximated by the convex functionals then the derivatives were used in the numerical method. In our case the original functional and its derivative are unknown, in addition they are non-convex.

3. ANALOGUE OF THE GRADIENT PROJECTION METHOD In this section we shall suppose that the admissible set

is a

convex compact in We shall use the sufficient conditions for the convergence of non-linear programming algorithms [21, 107]. L e m m a 1. Suppose that a sequence of points satisfies the following conditions: 1. is a compact set. 2. For an arbitrary convergent subsequence

the

following assumptions holds true: then a) if b) if

then there exists

such that

for any 3. There exists a continuous function values in is at most denumerable and following inequality

so that the set of its satisfies the

116

Chapter 3

Then the sequence the sequence

converges and all the limit points of belong to

Suppose that

where

Consider the following sequence of points

is an approximation of the performance criterion J(h), is a projective operator

is a step of the algorithm. Note that if value of

is a convex polyhedral set then calculation of a comes to a quadratic programming problem:

In the case when

there exist algorithms

which compute the value of the projective operator

in a finite

number of operations. T h e o r e m 1. Suppose that the sequence converges to J'(h) uniformly in the set The set of values of the function J(h) in the set

is at most denumerable, and

Then the limit of an arbitrary convergent subsequence of (7) belongs to the set


117

Note that for the convex functions the set is the set of minimum points of the function J(h). In the other cases the set contains all minimum points of the function J(h). The function J(h) that is under consideration can be non-convex because of non-linear state function dependence on the control h . To prove the theorem we shall apply Lemma 1. It is obvious that the conditions 1 and 2a of Lemma 1 are valid. Ensure that the other conditions of the lemma are true also, Suppose that there exists a subsequence Choose a sufficiently small such that a 4 å -neighbourhood of the point h´ and the set are mutually disjoint sets. By the definition of the set

there exists

Granting that

such that

for a sufficiently large k we have

Therefore, there exists

such that

Let us show that for all k > N we have sequence

where the

was defined in Lemma 1. Suppose the contrary. Then for all

thus

and

Define a function W(h) in the following way: W(h) = J(h). It is easy to see that the function W(h) satisfies all conditions of Lemma 1. Thus for all we have

118

Chapter 3

From continuity of the function J'(h), the inequality (8) and we obtain

Approaching the limit as in the inequality (10), we arrive at the inequality, which contradicts the boundedness of the continuous function W(h) on the compact set Therefore, By the construction,

so

for

sufficiently large and, hence, in the case the inequality (10) holds true by the same reasons as in the previous case. On the other hand,

whence

Substituting (11) into (10), we find that

Approaching the limit as

in (12), we obtain

Thus, all the conditions of Theorem 1 are valid, which is what had to be proved.


119

Remark 1. Since the a priori inequalities in the negative norms hold true, there exists a uniformly convergent to J'(h) sequence Remark 2. If the set of values of the function J(h) in the set is at most denumerable, it is possible to prove that there exists the subsequence (8), which is convergent to the set of solutions. Remark 3. An analytic function satisfies the condition of countability. Remark 4. There exists a function J(h) from such that the set of its value is at most denumerable in the set

4. ANALOGUE OF THE CONDITIONAL GRADIENT METHOD In the previous section applying the projective operation on the admissible set, we solved the optimization problem. Note that in some cases to execute the projective operation is very difficult. In this section we consider other approach to solving an optimization problem with constraints. Instead of projective operation the original optimization problem is substituted by a minimization problem of a linear function on the admissible set. Suppose that the conditions of Section 2 hold true and where

is a convex, closed and bounded admissible set. The

sequence of controls

where

is generated by the following algorithm:

is a step of the algorithm, which is chosen by the rule

120

Chapter 3

T h e o r e m 2. Suppose that the conditions of Theorem 1 are satisfied. Then the limit of an arbitrary convergent subsequence (13), (14) belongs to the set P r o o f. Apply Lemma 1. Conditions 1 and 2a hold true obviously. Check on the other conditions of Lemma 1. Suppose that there exists a subsequence Choose a sufficiently small such that the of the point h´ and the set are mutually disjoint sets. It follows from the definition of the set that for a sufficiently large k the following condition holds true

Since

as

uniformly on

Let us show that for all k = 0,1,..., the value Lemma 1 is finite. Suppose the contrary. Then, so that

whence we have

Suppose that W(h) = J(h) and

we have

defined in

for Observe that the function

J(h) satisfies all the conditions of Lemma 1. Then,

Since the first term becomes vanishingly small as

for a sufficiently large k we have

and


121

Taking into account continuity of the function J´(h) and the inequality

for a sufficiently small

we have

Considering inequalities (16) and (15), we obtain

contradiction with the boundedness of the continuous function W(h) as Thus, the sequence does not tend to infinity as By the definition of the sequence so

we have

for sufficiently large k

and, hence, in the case the inequality (17) holds true by the same reasons as in the previous case. Also, we have

whence

Substituting this inequality into (17), we obtain


we find

Thus all the conditions of Lemma 1 are valid. Remark. This approach based on Lemma 1 is very convenient for proving convergence of a non-linear programming algorithm

122

Chapter 3

in the case of programmed control of the step On the other hand, it should been noted that in the case of the step selection based on the complete step condition the assumption 2a of the Lemma is not valid. If proof of the assumption 2a is difficult or impossible at all, it is useful to employ the modification of Lemma 1 [108]. L e m m a 2. Suppose that a sequence of points satisfies the following conditions: 1. is a compact set, s = l,2,.... 2. If

then there exists

such that for all

we have 3. There exists a continuous function W(h) such that the set of its values in the set is at most denumerable and W(h) satisfies the following inequality

4. If

then

Under these assumptions the sequence an arbitrary limit point of the sequence

converges and belongs to set

Remark 2. If the set of values of the function J(h) in the set is at most denumerable, it is possible to prove that there exists a subsequence of (7), which is convergent to the set of solutions.


123

5. PROBLEMS OF JOINT OPTIMIZATION AND IDENTIFICATION 5.1 Formulation of joint optimization and identification problem There are a lot of optimal control problems described by a model with the vector a of unknown parameters. A sequence is an observation over this unknown vector a . In this case it is necessary to solve not only an optimization problem but also identification one. There exist a lot of approaches to solving this problem. One can first solve the identification problem with a given accuracy and then find the optimal control, nevertheless this approach possess an essential disadvantages: unknown parameters are estimated inaccurately, so that in solving the optimization problem the errors of control may be accumulated, in real time problems there are not enough observations for identification of parameters of unknown vector. The approach of joint optimization and identification is much more effective. This approach enables us to consider new optimization problems being directly related to limit extreme problems [25]. Although the above mentioned extreme problems are the problems of stochastic programming, it is necessary to develop special methods because the general methods of stochastic programming are not always effective. The following simple example shows one aspect of necessity to develop the special method of joint optimization and identification. It is requested to minimize positive quadratic form in the admissible set Assume that symmetric n×n matrix B is of the following form

124

Chapter 3

where is the positive defined (n – l)× (n –1) matrix. We have independent observations on an unknown parameter a . This problem can be solved in two ways. In the first case using the some observations we obtain an estimation of the parameter a , and then we begin to solve the deterministic problem. In the second case solving the identification problem we are making some iteration of the basic optimization algorithm using the current value of the estimate of the parameter a. The first approach is automatically inconvenient for real time problems. In addition, this approach can become incorrect. For example, if the identically distributed random variables s = 0,1,... take on the negative values much more frequently than the positive ones, then the estimate of a may be negative and the function may be non-convex. By

we mean

In particular, this is due to where p is probability,

Thus, possibility to solve the optimization problem by this approach essentially depends on the accuracy of the estimation of the parameter a. Conversely, the second approach is much more convenient for real time optimization problem. The consecutive approximations of the solution of the extreme problem are based on all currently observed


125

information about the parameter a. In this approach it is not necessary to find the vector a with high accuracy for concurrent execution of algorithms. In addition, the approach enables us to vary a method of step choice in the optimization algorithms. In this section we consider the problem

where a is a vector of unknown parameters. There are independent random observations The estimates

such that s =1,2,...

can be obtained in the same way as in the papers [109, 110]. To solve the optimization problem, we apply the algorithms similar to the well-known mathematical programming algorithms [21] of stochastic approximation. To prove the following assertion, we shall employ well-behaved sufficient conditions of convergence of stochastic programming algorithms. Following the paper [22], let us formulate the conditions. Let X be a set of solutions of some optimization problem, and be a random sequence of points.

T h e o r e m 1. Suppose that where s = 0,1,.... subsequences

for almost all

is a compact set. All convergent are satisfy the following conditions:

1) if 2) if that for all

then there exists a value

such where

3) there exists a continuous function W(x) such that the set of its values in the set X is at most denumerable and

126

Chapter 3

Under these assumptions the sequence

is

convergent and all the limit points of the sequence belong to the set X for almost all In the case when to verify the condition 1) is difficult, it is convenient to employ other conditions of convergence. T h e o r e m 2. Assume that for almost all the sequence belongs to the compact set All convergent subsequence

satisfy the following conditions:

1) if

then we

have

where

2) there exists a continuous function W(x) such that the set of its values in the set X is at most denumerable and

3) if

then

for almost all Under these assumptions the sequence convergent and all limit points of the sequence

is belong

to the set X for almost all This theorem is proved in the same manner as the previous one.


127

5.2 Analogue of the generalized gradient projection method Assume that a point

belongs to the convex compact admissible

set of the problem (1), (2). For all a the function is convex with respect to the variable x. A sequence of points is generated by the algorithm where

is a generalized gradient of the function F(a,x), s is

a number of iteration, is an initial approximation, is a projective operator to the set X; is iteration step of the algorithm. The parameter a is determined by the rule

T h e o r e m 3. Let F(a,x) be a continuous separately with respect to a and x,

Then with probability 1 the limit of an arbitrary convergent subsequence of (3) belongs to the set of the solutions X of the problem (1), (2), where P r o o f . Apply Theorem 1. According to the algorithm, the sequence belongs to the compact set X, s = 0,1,.... Henceforth, the dependence on shall be omitted. The first condition of Theorem 1 is obvious

128

Chapter 3

Verify the other conditions of Theorem 1. Let arbitrary convergent to a point

be an

subsequence. Then

Since the function F(a,x) is continuous and

with

probability 1, from inequality for a sufficiently large s, we have

From convexity of the function F(a, x) we have

Show that

for all k. Suppose the contrary. Then

for all

so that

that

for and

From the inequalities (5) and

From this it follows

all

Let

then

for all

have Whence, for sufficiently large k we obtain the inequality

we


129

Summing up the inequality over all the values of s from

to

r – 1, we have

Passing to the limit as in inequality (6), we obtain the contradiction with the non-negativity of the function W(x). Therefore,

It is clear that large

then for all sufficiently and hence in the case

inequality (6) is

true by the same reasons as in the previous case. On the other hand,

whence

Substituting this inequality into (6), we find that


in the inequality (7), we obtain

Thus, all conditions of Theorem 1 hold true, which is what had to be proved. Remark. If in Theorem 3 the function F(a,x) is continuously differentiable but non-convex, then an arbitrary convergent subsequence of the sequence (3) converges to points of a set

with probability 1.

130

Chapter 3

5.3 Analogue of the conditional gradient method In this section to solve the problem (1), (2) we consider a sequence

The parameter a is identified in the same manner as in the case discussed above

By a solution of the problem we mean the set X which is defined in previous remark. T h e o r e m 4. Let be a continuously differentiable function for all a and be continuous one for all The set of values of the function F(a, x) in set X is at most denumerable and

Then a limit of an arbitrary convergent subsequence belongs to the solution set X with probability 1. P r o o f . The first condition of Theorem 1 holds true obviously. Assume that there exists a subsequence Choose a sufficiently small such that the of the point x´ and the set X are mutually disjoint. By the definition of the set X, we obtain the inequality for sufficiently large k.


Suppose the inequality Whence, we have

holds true for all for sufficiently large k. Thus,

the inequality (11) holds true for

Assume that

the continuous function W(x) = F(a,x) and

Since

131

Consider Then

and

we have

for sufficiently large k. Taking into account continuity of the function inequality

for a sufficiently small

and

we obtain

Considering the equalities (12) and (13), we find

Passing to the limit as we obtain that the function W(x ) is unbounded, contrary to continuity of the function W(x) on the compact set X. Thus, From

we have

for

sufficiently large k. Then, in the case the inequality (14) holds true by the same reasons as in the previous case. On the other hand,

132

Chapter 3

whence,

Substituting this inequality into (14), we find that


we obtain

Thus, all conditions of Theorem 1 hold true, which is what had to be proved. Remark. If the set of values of the function F(a, x) in the set X is at most denumerable, it is possible to prove that there exists a convergent to the set of solutions subsequence of (8).

6. GENERAL THEOREMS OF GENERALIZED OPTIMAL CONTROL In the following sections we shall prove different a priori inequalities in the negative norms for different distributed systems, so the general optimal control theorems of the previous chapter hold true for this distributed systems. It is clear that the theorems from different section have common structure. We shall not formulate all theorems entirely in each section. We shall formulate only the changeable part of these theorems. Therefore in this section we collect only templates of the theorems of Chapters 1 and 2 in the convenient for the following sections form. Suppose the state function satisfies the following equation


133

with initial and boundary conditions (bd). (2) From the general theorems of Section 1.2 we conclude that the following theorems hold true.

T h e o r e m 1. Consider the problem of optimal control (1), (2). If 1) performance criterion is weakly lower semicontinuous; 2) the admissible set is bounded, closed, convex in H; 3) H is a reflexive Banach space; 4) is a weakly continuous mapping of the space H into 5) the operator L, the spaces N and from the following table

are chosen

then there exists an optimal control of the system (1), (2). T h e o r e m 2. Consider the system (1), (2) with the righthand side If the space and mapping

are chosen from the table below then the mapping is weakly continuous

Consider the question of the differential properties of performance criterion

134

where

Chapter 3

are some functions from

and

respectively, and Taking into account the general theorem from paragraph 1.4 and inequalities in the negative norms, we have T h e o r e m 3. Consider the problem (1), (2) with the righthand side If there exists a Fréchet derivative of a

mapping

at the

certain point h * , then at the point h* there exists a Fréchet derivative of the performance criterion J (h) in the following form

where v(t,x) is a solution of the adjoint problem

bilinear form If in addition

is defined in the spaces Fréchet

derivative

is

continuous at the point h* or satisfies the Lipschitz condition with index in a bounded and convex neighbourhood of * the point h , then the gradient has the same properties (is continuous or satisfies the Lipschitz condition with index where operator L, spaces are chosen form the following table

P r o o f is followed from the theorems of 2.1 (Theorems 2.1.32.1.5). T h e o r e m 4. Consider the problem (1), (2) with the righthand side There exists Fréchet a derivative


135

of the mapping space and the mapping following table.

if the are chosen from the

The Fréchet derivative

satisfies the

Lipschitz condition with index Consider a regularized problem of optimal control where It is requested to minimize a functional The functional is defined on functions from T h e o r e m 5. Consider the optimal control problems (1), (2) and (3), (2). Let the following conditions hold true: 1) admissible set of control is convex, closed, and bounded in the Hilbert space H ; 2) maps satisfy the condition:

a)

in for an arbitrary sequence

b) c)

is weakly continuous, for

all

fixed

d) the performance criterion is upper semicontinuous and weakly lower semicontinuous.

136

Chapter 3

Under these assumptions there exist optimal controls h*, of the problems (1), (2) and (3), (2); there exists a weakly convergent subsequence and an arbitrary weakly convergent subsequence

converges to h* weakly in H , where the

operator of initial boundary problem L , the space chosen from the table

are

T h e o r e m 6. Consider the regularized right-hand side of the equation (1), (2) from Section 2.2. If the operator L, the space the map are chosen from the following table, then conditions 2a)-2d) of previous theorem hold true for the maps

T h e o r e m 7. Consider the optimal control problems (3), (2) with the right-hand size If the space the exponent following table, then derivative

the map there exists of the map

are chosen from the a Fréchet derivative the

is defined in Section 2.2 and

satisfies the Lipschitz condition with index and directional derivative with respect to satisfies the Lipschitz condition with index

Chapter 4 PARABOLIC SYSTEMS

1.

GENERALIZED SOLVABILITY OF PARABOLIC SYSTEMS

Consider applications of the results of Chapters 1 and 2 to investigation of systems governed by parabolic partial differential equations. Note that despite of enormous number of papers devoted to the parabolic systems there are many open problems such as the problems of singular control and the problems of coefficient control. Let the functioning of a system is described by the following equation:

where u(t, x) is a function describing the system state in a region is a bounded region of n -dimensional Euclidian space Let

with a smooth boundary be functions which are continuously

differentiable in a closed region function such that

and c(x) is a continuous in

where the constant Denote by

a completion of the space of smooth functions

which satisfy the initial and boundary conditions with respect to the Sobolev norm

138

Chapter 4

is a similar space but its smooth functions satisfy the boundary conditions of the adjoint problem By pairs of the spaces construct negative spaces

and and

we

as a completion of the space

of smooth in functions satisfying the conditions (2) or (3) with respect to the norm

Let us study the properties of the differential operator (1), and also the properties of the formally adjoint operator:

L e m m a 1. For any function u(t, x), which is smooth in and satisfies conditions (2), the following estimation is valid: P r o o f . The inequality is obtained as a result of applying of the formula of integration by parts and the integral Cauchy inequality to the expression which is written in the right-hand side of definition of the negative norm. Indeed, let v(t, x) is a smooth in function satisfying conditions (3). Then

PARABOLIC SYSTEMS

139

where

Taking into account the integral representation of the function u(t, x) satisfying the initial condition we have

Applying to

the following inequality

140

Chapter 4

we have

Applying to the first term in the Ostrogradski-Gauss formula and taking into account the boundary conditions we conclude that this integral is equal to zero. Applying the integral Cauchy inequality to the second term in and taking into account that the functions are continuous, and hence they are bounded, we get

Let us apply to the third and fourth terms (5). We get

the inequality like

Finally, we have Hence, the statement of lemma holds true for smooth functions u(t, x) as far as the set of considered functions v(t, x) is dense in and hence the supremum in the definition of the negative norm we can take at such smooth functions. Re m a r k . The proved inequality allows extension of the operator L with respect to continuity on the whole space in this case the inequality of the Lemma still valid for any function u(t, x) of the space

PARABOLIC SYSTEMS

L e m m a 2. For any function

141

the

following

estimation is valid The P r o o f of Lemma 2 is similar to the proof of Lemma 1. L e m m a 3. For any function the following estimation is valid P r o o f . Consider the following auxiliary integral operator defined on smooth functions u(t, X) , which satisfy conditions (2):

where

is a constant majoring

which exists by virtue the

fact that the functions are continuous in n is the dimension of the region It is clear that the function v(t, x) satisfies conditions (3). Expressing u(t, x) via v(t, x) we get

Consider

where

142

Consider every term

Chapter 4

and

separately.

Using the Ostrogradsky-Gauss formula and taking into account the condition we can prove that the integral is equal to zero. Let us apply to the second term the formula of integration by parts:

Using the Ostrogradsky-Gauss formula we pass to the surface integral in the term account the condition inequality

and take into Applying to the second term the we have

PARABOLIC SYSTEMS

143

Consider the term

Consider the last term

Applying the formula of intergation by

parts and taking into account the condition

we get

144

Chapter 4

Finally, we obtain

Applying the Schwarz inequality to

we obtain

Reducing the right-hand side and the left-hand side on the inequality by and taking into account the relation between u(t, x) and v(t, x), we obtain

The validity of the inequality (7) on the whole space we prove by passing to the limit. L e m m a 4. For any functions the following inequality holds true: The P r o o f of Lemma 4 is carried out in a similar way as for Lemma 3. The auxiliary operator is of the form

PARABOLIC SYSTEMS

145

T h e o r e m 1. For any function there exists a unique solution of the problem (1), (2) in the sense of Definition 1.1.1. T h e o r e m 2. For any functional then a unique solution of the problem (1), (2) exists in the sense of Definition 1.1.4.

2. ANALOGUE OF GALERKIN METHOD Let us construct the method of numerical solving the parabolic equation with a generalized right-hand side. Consider the following equation

where u(t, x) is a system state defined on the set

Let the system state is described by the initial-boundary value problem (1), (2), We shall look for the approximate solution in the following form

146

where

Chapter 4

is a orthonormal basis in

functions from the space

which consists of

and the functions

are selected

as a solutions of the Cauchy problem for the system of N linear ordinary equations:

L e m m a 1 . The following inequality is valid

P r o o f . Multiplying both parts of expression (5) by and summing with respect to l from 1 to N and integrating with respect to t from 0 to T , we obtain

PARABOLIC SYSTEMS

147

Integrating by parts and using the initial and boundary conditions, we get the following expressions

Using the Ostrogradsky-Gauss formula we obtain that the first integral in the right-hand side equals to

Let us

apply to the second integral the integration by parts. Thus, we have

Applying the Ostrogradsky-Gauss formula again and taking into account

the

coercitivity we obtain

condition:

148

The term

Chapter 4

we estimate in the following way.

Let us apply to the term

the formula of integration by parts.

Summing the obtained expressions, we have

PARABOLIC SYSTEMS

149

Comparing the obtained expressions with (6) and using the integral Cauchy inequality, we obtain

Reducing both parts by

we finish to prove Lemma 1.

T h e o r e m 1. Let

The

sequence

of

approximations (4) converges to the solution u(t, x) of the problem (1), (2) in the sense of Definition 1.1.1 with respect to the norm of the space P r o o f . By Lemma 1, we can extract the weakly convergent subsequence from the bounded sequence Let be weakly convergent to

By the Banach-Sax

theorem, we can extract from it a subsequence strongly converges in

such that

to the same function û , i.e.

150

Chapter 4

By virtue of the compactness of the imbedding

it is

strongly convergent in Taking into account the inequality of Lemma 2.1.1 and the linearity of the operator L , we get

By the fact that sequence

is fundamental, we have

As far as the space

is complete the fundamental sequence

has a limit

i.e.

Remark. Element and

does

not

Let us prove that

is identically defined by depend

on

the

sequence

in the sense of the equality in the space

Let us multiply both parts of the equality (5) on an arbitrary smooth function

which satisfies the condition

integrate it with respect to t from 0 to T. Denoting we obtain

and

PARABOLIC SYSTEMS

151

On the basis of the definition of the function equalities hold true

the following

On the other hand, by the Schwarz inequality we have

By virtue of the above reasoning, the right-hand side of this inequality tends to zero as That is why it follows from (7) that We can make the number

be arbitrarily large (dropping the

necessary of the first terms of the sequence and repeating the analogous reasoning). In view of the fact that the set of functions we have in whence it is easy to prove that û is the solution of the problem (1), (2) in the sense of definition 1.1.1. By virtue of the uniqueness of the solution of the problem (1), (2), all sequence converges to û with respect to the norm Let us consider the case when the right-hand side of the equation (1) is an element of the negative Hilbert space Granting the density of the space

let us select a sequence (we can do it applying the well-

known procedure of averaging). Consider the problem (1), (2) with the right-hand side shall look for the approximate solution of this problem in the form

We

152

Chapter 4

where

is above defined orthonormal basis in

and

is determined from the solution of the Cauchy problem for the system of linear ordinary differential equations:

As far as the estimations in the negative norms are valid for the operator L passing to the limit we shall obtain that the following statement holds true T h e o r e m 2. Let Then the sequence obtained by the method (9), (10) with the help of the special choosing of N = N(p), which necessarily exists, converges to the generalized solution from the space (in the sense of Definition 1.1.4) of the problem (1), (2) in the following sense P r o o f.

By Theorem 1, sequences of the approximations converge to the generalized solutions of the problems

with respect to the norm

i.e.

PARABOLIC SYSTEMS

153

Let us show that the sequence space

is fundamental in the

We have

Next,

Here we used the estimation of Lemma 1 and the triangle inequality. Let us pass to the limit with respect to We shall obtain that Hence, there exists a function

such that

It follows

from the proof of Theorem 1.1.3 that the function is a generalized solution of the problem (1), (2) in the sense of Definition 1.1.4 with the right-hand side Consider

On the basis of Theorem 1 we can

choose N = N(p) such that

where

is a sequence of positive numbers converging to zero. Hence,

The theorem is proved. Thus, in the case when the right-hand side belongs to the negative space it is necessary to solve the Cauchy problem (10) and to co-ordinate the parameters N and p in a special manner.

154

Chapter 4

3. PULSE OPTIMAL CONTROL OF PARABOLIC SYSTEMS Let us consider the problems connected with the pulse optimal control in the case when the function of the system state is a solution of the Dirichlet initial-boundary value problem for parabolic equation:

All notations correspond to Chapter 1, where we proved the inequalities in the negative norms for the parabolic operator. Using the templates of the theorems mentioned in Chapter 3, let us write the tables for this theorems.

PARABOLIC SYSTEMS

155

Let us investigate the problem of the existence of the optimal control of coefficients of the equation. Let the system state is described by the following equation:

156

Chapter 4

where u(t, x) is a function describing the system state in the domain is a bounded domain in n dimensioned Euclidian space with smooth boundary which is a tube with respect to the variable The control of the system is carried out with the help of control impacts The functional which is to be minimized on the set of admissible controls, is defined on the solutions of the problem (3) and is weakly lower semi-continuous with respect to the system state u(t, x). Let be continuously differentiable function in the closed domain

and

shall assume that

be continuous function in

We

are some numerical parameters and

for all values of this parameters

where the constant Let us consider the specific case of the control impact which arise in practice and investigate the problem of existence of the optimal controls. Let the following conditions are satisfied:

PARABOLIC SYSTEMS

where the vector

157

is a control.

Then the following statement holds true. T h e o r e m 10. Let the system state is determined as a solution of the problem (3), (2) and the following conditions are satisfied: 1) the performance criterion is a weakly lower semi-continuous with respect to the system state functional; 2) the set of admissible controls is bounded, closed and convex in H; 3) H is a reflexive Banach space; 4) f is a weakly continuous operator mapping H into Under these assumptions there exists an optimal control of system (3), (2). P r o o f . Note that by virtue of the relations (4), (5) all statements proved in the Section 1 for the parabolic operator are valid for differential operators (3). Moreover, the constant C in the estimations with respect to the negative norms (Lemmas 1.1-1.4) does not depend on the control Hence, the proof of Theorem 1 is completely similar to the proof of the general theorem about existence of optimal control (Theorem 1.2.1)

158

Chapter 4

Remark. As far as the operator f is non-linear then the functional J(h) may be non-convex and the optimal control may be non-unique. Let the functioning of the system is described by the equation (3) with the conditions (4), (5). Let us investigate the differential properties of the performance criterion

where

are some unknown functions from

and respectively, and the optimal control of the system coefficients. Let us give to the control h an increment

in

for the case of

where Let us introduce the adjoint state as a solution of the following problem

Then we may write the increment of the performance criterion in the following form:

PARABOLIC SYSTEMS

159

As far as then, obviously,

satisfies the equation

It follows from the results of Section 1.1 that the solution of this problem exists, it is unique and it is determined as a function such that for any (including v(t, x)), the following equality holds true:

Thus,

160

Dividing both parts by have

Chapter 4

and passing to the limit as

we

Thus, we have proved the following statement.

T h e o r e m 11. The gradient of the performance criterion (6) is of the following form: gradJ (h) =

Remark. It should been noted that the passing to the limit in (8) was done formally, but its justification can be proved easily.

Chapter 5 PSEUDO-PARABOLIC SYSTEMS

In order to solve many applied problems of science and engineering it is necessary to study optimal control problems for pseudo-parabolic systems: where A, B are second order elliptic differential operators. Seemingly one of the first paper on pseudo-parabolic equation is [111], where the equation was obtained from the research of heat transport processes in the heterogeneous environment, as more adequate model of the processes. Pseudo-parabolic equations arise in researches of the fluid and gas filtration in the fissured and porous medium [112-116]; the heat conduction in the heterogeneous environment [111], the ion migration in soil [111-113, 117], the wave propagation in the disperse medium and in the thin elastic glass [118]. Nowadays , there are many papers on pseudo-parabolic equations [51, 52,90,119-128].

1. GENERALIZED SOLVABILITY OF PSEUDOPARABOLIC EQUATIONS (THE DIRICHLET INITIAL BOUNDARY PROBLEM) Consider equation (1) in a tube

where

regular domain in with a piecewise-smooth boundary operators A,B are defined as

is a The

162

Chapter 5

where

are

continuously differentiable functions in the closed domain and a(x), b(x) are continuous functions in the We suppose that the differential expression (2a) is positive definite and the differential expression (2b) is nonnegative in the domain i.e.

where

is a positive constant,

We also suppose that Introduce into consideration the following spaces. Let

be a

completion of the set of the smooth functions in the domain satisfy the conditions

which

in the norm

Let domain

be a completion of the set of the smooth functions in the which satisfy the adjoint conditions

in the same norm (5);

are corresponding negative spaces Let

completion of the set of the smooth functions in the domain satisfy the conditions (4) by the norm

be a

which

PSEUDO-PARABOLIC SYSTEMS

163

be completion of the set of the smooth functions in the domain which satisfy the conditions (6) in the same norm. Let be corresponding negative spaces.

L e m m a 1. For all functions inequality is true

the following

P r o o f. First consider a smooth function u(t, x), that satisfies the conditions (4), then applying expansion by continuity of operator and passing to the limit, we shall obtain the inequality for all functions By the definition of the negative norm, we have

Consider

Employing the integration by parts, the

Schwarz inequality and the conditions (6), we obtain

Here we apply the inequality

In the same manner, we have

Using the integration by parts, we find

164

Chapter 5

Passing to the integration over surface and taking into account the. conditions (6), we conclude that

Using the Schwarz inequality again, we obtain

Analogously

Substituting the inequality into the parity (7), we have the desired inequality. The analogous lemma is valid for the adjoint operator L e m m a 2. For all functions the following inequality is true

where It follows from these lemmas that the operator L (and can be extended to the continuous operator mapping (and respectively) into L e m m a 3. For all functions inequality is true

the following


165

P r o o f. First consider a smooth function u(t,x), which satisfies conditions (4). Let v(t, x) be an auxiliary function in the form

It is obvious that

Prove the following inequality

Using the integration by parts, the relationship between u(t, x) and v(t, x), the boundary conditions, we have

Next, in much the same manner, we have

Employing the integration by parts, we have

166

Chapter 5

Passing to the surface integral, since the symmetry and nonnegativity of the matrix we arrive

Substituting (10) into (9) from the obtained inequalities, we have

whence, applying the obvious inequality

we conclude that the inequality (8) is valid. By the Schwarz inequality, we have

Reduce by

and take account the relationship between

u(t,x) and v(t,x). Whence we have that desired inequality of the lemma is true for a smooth function. Passing to the limit, we obtain the inequality for all functions L e m m a 4. For any function inequality holds true

the following


Proof. Let first

167

be a smooth function, which

satisfies conditions (6). Prove the following inequality where

Using the integration by parts and the definition of the function u(t,x),we have

Since the functions u(t,x) and v(x,t) satisfy homogeneous conditions on the border of the set [0,T] , we have:

Next, we obtain

Since the matrix the inequality

is symmetric and nonnegative, we have

168

Chapter 5

Taking into account (12), (13) and the inequality

we obtain (11). Next, in much the same manner as in Lemma 3, applying the Schwarz inequality and passing to the limit, we prove desired inequality for all Based on Lemmas 1-4 and results of Chapter 1, we have the following theorems. T h e o r e m 1. For all there exists a unique solution of the problem (1), (4) in the sense of Definition 1.1.1. T h e o r e m 2. For all

there exists a unique

solution of the problem (1), (4) in the sense of Definition 1.1.4.

2. GENERALIZED SOLVABILITY OF PSEUDOPARABOLIC EQUATIONS (THE NEUMANN INITIAL BOUNDARY PROBLEM) In this section we shall consider the simpler (comparing to the general equation (1.1)) pseudo-parabolic equations. This simplification enables us to consider the initial boundary value problem with usual von Neumann conditions (compare with the initial boundary conditions of pseudo-hyperbolic systems). Consider the partial differential equation: in a tube

where u(t, x) is a sought function,

is a bounded domain in with a smooth border is a Laplacian, k is a nonnegative constant.


169

Together with the equation (1), consider the boundary conditions

where is a normal vector to the surface Introduce the following denotation. Let set of the smooth in norm

be a completion of the

functions, which satisfy the condition (2) in the

is an analogous space, but functions satisfy the adjoint conditions

are corresponding negative spaces. L e m m a 1. For all functions inequality is true P r o o f.

the following

First prove the lemma for smooth functions which satisfy the condition (2), and then passing to the

limit, we shall obtain the inequality for all functions By definition of the negative norm in

we have

170 Since the bilinear form with the inner product in

Chapter 5 on the smooth functions coincides we obtain

Consider

Employing the Schwarz inequality, we have

Let us show that

First consider the integral Using partial integration and the conditions (2), we find

Passing to the surface integral, we obtain


(since

and, thus,

171

).

Applying the Schwarz inequality to (5), we have

In the same manner, we prove that

Finally, we obtain

Returning to (4), we have as required. In the same manner, we prove the following lemma.

L e m m a 2. For all functions

the following

inequality holds true

where

is a formally adjoint operator

L e m m a 3. For all functions inequality is true

the following

172

Chapter 5

P r o o f. First prove the lemma for smooth functions u(t,x), which satisfy the condition (2), and then passing to the limit, we obtain the inequality for all functions Consider an auxiliary operator

Evaluate

Applying the integration by parts and the conditions (2) and (3), for functions u(t, x) and v (t, x) we have

Substituting the equality into (7) and replacing a(t) by – (t + 1), we have

From the Schwarz inequality, we obtain Reducing each part by

we find


which is what had to be proved. L e m m a 4. For all functions

173

the following

inequality holds true From Lemmas 1-4 we have the existence and uniqueness theorems.

T h e o r e m 1. For all functions there exists a unique solution of the problem (1), (2) in the sense of Definition 1.1.1. T h e o r e m 2. For all functions there exists a unique solution of the problem (1), (2) in the sense of Definition 1.1.4. 3. PULSE CONTROL OF PSEUDO-PARABOLIC SYSTEMS (THE DIRICHLET INITIAL BOUNDARY VALUE PROBLEM) Apply the obtained results for the optimization problem of pseudoparabolic systems (the Dirichlet initial boundary value problem). We use the same denotations as in Section 1. Let the state function satisfies the pseudo-parabolic equation

Using the template theorems of Section 3.6, fill in the tables for the pseudo-parabolic equation (the Dirichlet initial boundary value problem).

174

Table 4 is e m p t y.

Chapter 5


175

4. PULSE CONTROL OF PSEUDO-PARABOLIC SYSTEMS (THE NEUMANN INITIAL BOUNDARY VALUE PROBLEM) Apply the obtained results for the optimization problem of pseudoparabolic systems (the Neumann initial boundary value problem). We use the same denotations as in Section 2. Let the state function satisfies the pseudo-parabolic equation

176

Chapter 5

Using the template theorems of Section 3.6, fill in the tables for the pseudo-parabolic equation (the Neumann initial boundary value problem).

Table 4 is e m p t y.


177

Chapter 6 HYPERBOLIC SYSTEMS

1. GENERALIZED SOLVABILITY OF HYPERBOLIC SYSTEMS (THE DIRICHLET INITIAL BOUNDARY VALUE PROBLEM) This Section is devoted to thé research of hyperbolic partial differential equations. A lot of the mechanic and physics problems such as a chord, bar and membrane oscillation, an electromagnetic oscillation etc. are described by the hyperbolic equations [98, 129-133]. Consider a linear partial differential equation in a tube domain

is a bounded domain of

variation with a smooth domain boundary The elliptic operator B does not depend on the temporary variable t and is of the following form

Suppose

that

the

matrix

the matrix cells

is and the functions

continuously differentiable in the domain

symmetric are

and for all vectors

the following inequalities hold true

180

Chapter 6

Since the functions

are continuous on the compact set

there exists a constant

such that

for all

We denote by a completion of the set of smooth in functions satisfying the following conditions in the norm

Analogously let

be a completion of the set of sn.

functions satisfying the following conditions in the same norm

Let

be corresponding negative spaces.

L e m m a 1 . For all

and

the

following inequalities

hold true, where an operator L is formally adjoint to the operator L :

HYPERBOLIC SYSTEMS

181

P r o o f . Consider first a simpler case, when u(t, x) is a continuous function satisfying the conditions (2). By the negative norm definition, we have

since for smooth functions u(t, x) satisfying the condition (2), the bilinear form coincides with the inner product in Consider the term of the right-hand side of the fraction (4):

Applying the partial integration, the Schwarz inequality and the initial conditions (2), (3), we obtain

Next,

In the same manner, we have

182

Chapter 6

Returning to the equality (4), we obtain the lemma assertion for smooth functions u(t, x), which satisfy the condition (2). Passing to the limit, we obtain the lemma assertion for all functions The second inequality is proved in the same manner. L e m m a 2 . For all functions

and

the following inequalities hold true

P r o o f . Consider the first inequality. Suppose first that a function u(t, x) is smooth and satisfies the condition (2).Consider the following integral operator

where

is a positive constant.

It is obvious that

Consider

Estimate the every summand in the right-hand side. Applying the integration by parts, we have

Since

we obtain

HYPERBOLIC SYSTEMS

Taking into account that

183

and so

we arrive

The other summands of the equality (5) are estimated in the same manner as the previous one. Consider the second summand:

Next, estimate the third summand of (5):

184

Chapter 6

Examine the last summand of (5):

Applying the all acquired estimation, we have

Since

we arrive

HYPERBOLIC SYSTEMS

185

The term

can be made positive by choosing the constant

It suffices to put

We have

Thus,

Employing to the left-hand side the Schwarz inequality, we obtain

Dividing by

we arrive the following inequality:

Taking into account the definition of the function v(t, x), we have

186

Chapter 6

whence we finally obtain This inequality can be extended to any function by passing to limit. The second inequality is proved in the same manner. T h e o r e m 1. For all functions there exists a unique solution of the problem (1), (2) in the sense of Definition 1.1.1.

T h e o r e m 2. For all functions

there exists a

unique solution of the problem (1), (2) in the sense of Definition 1.1.4.

2. GENERALIZED SOLVABILITY OF HYPERBOLIC SYSTEMS (THE NEUMANN INITIAL BOUNDARY VALUE PROBLEM)

Suppose that a state function is described by the hyperbolic equation where B is an elliptic operator in the following form

the functions the domain

are continuously differentiable in The operator B is uniformly elliptic in

HYPERBOLIC SYSTEMS

The

equation

(1)

187

is

considered

Introduce the following denotations. Let the set of smooth in conditions

in

the

tube

domain

be a completion of

functions, which satisfy the following

in the norm

where

is a normal vector to the surface of the domain

at a point The boundary condition

can be written in the form

where vector

is the inner product in

is the following co-normal

wherå A is a matrix of elements Let

be a completion of the set of smooth in

which satisfy the following conditions

functions,

188

Chapter 6

in the norm (5). According to the pairs of sets

and

consider the negative spaces Prove a priori inequalities in the negative norms for a hyperbolic system. L e m m a 1 . For all functions the following inequality holds true P r o o f . Prove first the inequality for smooth functions u(t, x) , which satisfy the conditions (4). By the definition of the negative norm, we have

since for a smooth function u(t, x) the bilinear form the inner product

equals to

in the space

Consider the equality (7). Applying partial integration and the conditions (4), (6), we have

Adding the equalities and using the Schwarz inequality, we find

HYPERBOLIC SYSTEMS

Substituting the inequality into (7) and dividing by

189

we obtain

the desired inequality for all smooth functions u(t, x), which satisfy the condition (4). Considering the continuous extension of operator L on the space we have the desired inequality for all functions from L e m m a 2. For all functions

the following


where L is the adjoint operator. The p r o o f of Lemma 2 is completely analogous to the previous lemma. The inequalities from Lemma 1 and 2 show that the operators L and L map continuously the spaces and into and respectively. L e m m a 3. For all functions inequality holds true

the following

P r o o f . We first prove the lemma for a smooth function, which satisfies the conditions (4). Consider an auxiliary function v(t, x)

For all functions v(t, x) we shall prove the following inequality

190

Chapter 6

Consider the inner product Employing the integration by parts and relation between the function u(t, x) and v(t, x), we have

Next, in the same manner, we have

where Taking into account symmetry of the matrix integration by parts, we find

Substituting the inequality into (10), we arrive

From (9) and (11) we have inequality (8).

and applying

HYPERBOLIC SYSTEMS

191

Employing the Schwarz inequality to the left-hand side of (8), we obtain Comparing formulae (8) and (12), dividing by

and taking into

account the relation between u(t, x) and v(t, x), we have the assertion of Lemma 3 for a smooth function u(t, x), which satisfies the condition (4). Passing to the limit, we obtain the assertion for all functions L e m m a 4. For all functions

the following

inequality holds true P r o o f . Let first v(t, x) be a smooth function that satisfies the condition (6). Consider the following auxiliary function u(t, x):

Show the inequality Applying the integration by parts, relation between u(t, x) and v(t, x), we have

In the same manner, consider the other summands of

192

Chapter 6

Substituting (14) and (15) into (13), we obtain the assertion of Lemma 4 for a smooth function Passing to the limit, we obtain the assertion for all functions Applying the theorems of Section 1.1, we have the following theorems for the problem (1), (4). T h e o r e m 1. For all functions there exists a unique solution of the problem (1), (4) by Definition 1.1.1. T h e o r e m 2. For all functions there exists an unique solution of the problem (1), (4) by Definition 1.1.4.

3. GALERKIN METHOD FOR HYPERBOLIC SYSTEMS In this section we consider the application of the Galerkin method for approximate solving the hyperbolic equation (the Dirichlet initial boundary value problem). We shall find the approximate solution of the equation

in the form

HYPERBOLIC SYSTEMS

193

where

is an orthonormalized basis of the space

The

functions

satisfy the following conditions

are chosen as a solution of the Cauchy problem for the following set of the linear ordinary differential equations:

From the set of equations (3) it follows that

L e m m a 1. The following inequality holds true P r o o f . Multiplying both the left and right hand sides of the equation (4) by

where the constant

was defined in

Section 1, and summing up over j from 1 to k, we obtain:

194

Chapter 6

Applying the integration by parts, passing to the surface integration and taking account the boundary conditions

we find

Next, in the same manner, we have

Since

and thus

we conclude that first

integral of the right-hand side is equal to zero. Consider the second integral of the right-hand side:

HYPERBOLIC SYSTEMS

Consider the third and fourth summands of (5):

Summing up the obtained inequalities, we have

195

196

Chapter 6

Taking into account the value of the constant that

Thus,

By the Schwarz inequality, we have

we obtain

HYPERBOLIC SYSTEMS

Dividing by

we obtain the desired inequality.

From the lemma it follows that the sequence the space subsequence

197

is bounded in

so there exists a weakly convergent to To prove the strong convergence, consider

the following spaces. Let be a completion of the set of smooth in which satisfy boundary conditions in the norm

functions,

Denote by and completions of the set of smooth in functions, which satisfy conditions (2), in the following inner products:

respectively. Integrating by parts and passing to the surface integration, it is easy to prove the following lemma. L e m m a 2. For all smooth in functions u(t, x), which satisfy the conditions (2), the following inequality holds true

198

Assume that function

Chapter 6

f (t, x) satisfies conditions

and

Differentiating the both right-hand and left-hand sides of (3) with respect to t , we obtain

Multiplying the both right-hand and left-hand sides by summing up over j from 1 to k and integrating over t from 0 to T, we have

Consider the integrands of the left-hand side of (6).

We have Next, we have

by choosing the function f(t,x).

HYPERBOLIC SYSTEMS

199

In the same manner as in Lemma 1, consider the other integrands.

Summing up the obtained inequalities, we have

200

Chapter 6

Taking into account the value of the constant

we

obtain

Since (6), we conclude that

Applying to the right-hand side the Schwartz inequality, we find

Thus, the sequence

is a weakly compact set in the spaces

so there exists weakly convergent to subsequence. From (7), we have

HYPERBOLIC SYSTEMS

201

Whence, there exists a weakly convergent to subsequence

In the same manner the weak convergence of the

other partial derivative sequences may be proved. Thus, we obtain

Observe that for all functions

where

is the bilinear form being generated in the triple of Hilbert

spaces Since the functions

satisfy equality

rewrite the relation (8) in the following form:

On the other hand, from the equality (3) we have that right-hand side of (9) equals to zero. Thus, the sequence and, hence, in space

is strong convergent to

in the space

From Lemma 1.1 it follows that the

202

Chapter 6

sequence

is fundamental in the complete space

so

Multiplying the both right-hand and left-hand sides of (4) by a smooth function g(t) (g(T) = 0) and integrating over t from 0 to T, we obtain: Approaching the limit as

Since the set in the space

we have

is total in the space thus the function

we conclude that is a solution of

the problem in the sense of Definition 1.1.2 and, hence, in the sense of Definition 1.1.1. We note that there was no necessity to choose a subsequence because of uniqueness of the solution. The obtained result we formulate in the following theorem. T h e o r e m 1. For a function the approximate sequence (3) converges to the solution of the problem (1), (2) in the sense of Definition 1.1.1 in the norm of the space Taking into account the density of the considered functions f(t,x) in the space it is easy to prove the following theorem. T h e o r e m 2. For all functions

the approximate

sequence (3) converges to the solution of problem (1), (2) in the sense of Definition 1.1.1 in the norm of space Consider theapplication of the Galerkin method when the right-hand side of the state equation belongs to the negative space

HYPERBOLIC SYSTEMS

Let

203

be an arbitrary function from

density of the space

in the space

sequence of functions

By virtue of the

there exists a

such that

In this case, we consider the approximate sequence in the following form: where

the function

is a solution of the following set of ordinary

differential equations

By Theorem 2 the sequence of the problem sequence

converges to the solution

in the norm of space

Consider the

By the Lemma 1.1.3, we have

Thus the following theorem holds true. T h e o r e m 3. For all functions

the

approximate sequence (11) converges to the solution of the problem (1), (2) in the sense of Definition 1.1.4 as in the norm of space

204

Chapter 6

4. PULSE OPTIMAL CONTROL OF HYPERBOLIC SYSTEMS (THE DIRICHLET INITIAL BOUNDARY VALUE PROBLEM) Using the template theorems from Section 3.6, we shall fill in the following tables.

HYPERBOLIC SYSTEMS

205

Chapter 7 PSEUDO-HYPERBOLIC SYSTEMS

1. GENERALIZED SOLVABILITY OF PSEUDOHYPERBOLIC SYSTEMS (THE DIRICHLE INITIAL BOUNDARY VALUE PROBLEM) 1.1 Formulation of the problem. Main notations. In this chapter we shall consider the problems of optimization of the systems described by the pseudo-hyperbolic equations. Such equations arise, for example, in the investigations of mass transport in heterogeneous porous media [134]. In addition, many processes are described by non-stationary equations with small viscosity. For instance, torsion oscillations of metallic cylinder with inner friction, propagation of perturbations in viscous and elastic rod, one-dimensional flow of isotropic viscous liquid, sound propagation in viscous gas and similar processes are described by the model equations in the following form:

where

is a parameter,

is a small viscosity.

There are well-known the equations of viscous and elastic medium [135]

where

are the Lame constants

are the viscosity constants, is density. For example, the equations of longitudinal oscillations of viscous and elastic rod has the following form [136]:

208

where

Chapter 7

is density of the medium, E is the

elasticity coefficient, is the viscosity coefficient. The equation of the propagation of initial perturbations in viscous gas has the following form:

where C is the sound velocity in the absence of viscosity, is the cinematic coefficient of viscosity. Pseudo-hyperbolic systems were investigated in papers [137140,172,176]. Let us consider the system with distributed parameters and pseudohyperbolic state equation. P r o b l e m 1 . Find a function u(t,x) satisfying the following equation

where u(t, x) is defined in the domain is a bounded domain with a sufficiently smooth bound Suppose that


are continuously differentiable in the closed domain

209

functions,


The adjoint equation has the following form

Let the system state satisfies the following conditions:

210

Chapter 7

Let us introduce the following notations. is a completion of the space of smooth functions, which satisfy the conditions (5), in the norm

is the same space, but the functions satisfy the following conditions

are the analogous spaces obtained by completing of smooth functions satisfying the conditions (5) and (7), respectively, in the norm

are the correspondent negative spaces. The following imbeddings are valid: and, moreover, imbeddings are dense and the imbedding operators are completely continuous. P r o b l e m 2. Consider the same equation when its coefficients do not depend on time


211

where u(t, x) is defined in the domain is a bounded domain with a sufficiently smooth bound Suppose that


functions; a(x ), b(x) are continuous in

functions;

In Problem 2 we shall consider the same boundary conditions and chains of Hilbert spaces as in Problem 1.

P r o b l e m 3 . We shall investigate the equation

212

Chapter 7

where u(t, x) is defined in the domain bounded domain with sufficiently smooth bound Suppose that

is a are

continuously differentiable in the c1osed domain functions, M = M(x) is continuous in the closed domain function;

where

are positive constants.

Let us introduce the following denotations: is the completion of the space of smooth functions, which satisfy the condition (5), in the norm

is the same space, but the functions satisfy the conditions of the adjoint problem (7); spaces.

are the corresponding negative

1.2 Properties of pseudo-hyperbolic operator in Hilbert space Let us investigate the properties of the operators first, we shall show that the operator

and

At

is extendable to the operator,

which continuously maps the whole space

into space

.


213

L e m m a 1 . For the functions priori inequalities holds true:

the following a

where C hereinafter is some positive constant. P r o o f . At first, we shall prove the lemma for smooth functions u(t, x), which satisfy the condition (5), and later we shall obtain the validity of the lemma for any extending in the correspondent way the operator and passing to the limit. By definition of the negative norm, we have

as far as for smooth u(t, x), which satisfy the condition (5), the bilinear form coincide with the inner product in consider the numerator in the right-hand side of (11):

Let us

Using the integration by parts, the integral Cauchy inequality and granting the initial conditions, we can write

214

Next,

In a similar way,

Chapter 7


215

From the written above we obtain

Using obtained inequality, we have

for smooth functions u(t, x), which satisfy the condition (5). Using the inequality (10), we can extend the operator which is defined on smooth functions satisfying the condition (5) to the whole space (continuity extension). For extended operator we shall save the previous notation and hereinafter we shall consider only extended operator. Passing to the limit, we obtain the validity of the lemma for any functions

216

Chapter 7

Remark 1. The inequality (10) imply that the linear operator continuously maps the whole space into

In a similar way we can prove that extended operator continuously maps the whole space

into the space

L e m m a 2 . For all functions

the following


Statement 1 .

Let

then the following

equality holds true For Problem 2 the following lemmas hold true. L e m m a 3 . For all functions priori inequality holds true:

the following a

The proof of Lemma 3 is analogous to the proof of Lemma 1. L e m m a 4. For all functions the following inequality holds true The proof of Lemma 4 is similar to the proof of Lemma 2. For Problem 3 the following lemmas hold true. L e m m a 5 . For all functions the following a priori inequality holds true The proof of Lemma 5 is similar to the proof of Lemma 1.


L e m m a 6 . For all functions

217

the following


The proof of Lemma 6 is similar to the proof of Lemma 2.

1.3. The existence and uniqueness of solution of initial boundary value problem To investigate the solvability of an initial boundary value problem we shall use equipped Hilbert spaces, a priori estimations in negative norms, and modified method of deriving of energetic inequalities in negative spaces. P r o b l e m 4 . We shall investigate the solvability of the following problems

are given elements. where u, v are unknown elements, and The solutions of the problems (12), (13) we shall consider as generalized solutions in the following sense. D e f i n i t i o n 1 . The generalized solution of the problem (12) is a function smooth functions

such that there exists a sequence of which satisfy the condition (5) and

D e f i n i t i o n 2 . The generalized solution of the problem (12) is a function

such that the identity

218

Chapter 7

holds true for any smooth functions v(t,x) satisfying the condition (7). In a similar way we can introduce the definition of a generalized solution of the adjoint problem.

L e m m a 7 . For any function inequality holds true

the following

P r o o f . At first, let us prove the inequality for smooth functions u(t, x) satisfying the condition (5). Introduce an auxiliary function v (t, x) in the following way:

where

The definition of the function V(t, X) implies that Note that

Let us prove the validity of the following inequality: Consider


219

Using the integration by parts, relation between u(t, x) and V(t, X), and initial conditions, we obtain that

Granting initial and boundary conditions, we have

In a similar way,

220

Granting the fact that the matrix

Chapter 7

is symmetric and

nonnegative we conclude that the following relation is valid:

Next ,


221

Note that it is possible to prove the convergence of the improper integrals, which are the terms of the sums. Thus,

The proved relations imply that (15) holds true. Applying the Schwarz inequality to (15), we obtain:

Reducing by

we have the following inequality:

Next, let us consider

Applying to the right-hand side the theorem of the mean, we have

222

Chapter 7

where θ is some mean value from the interval (0, T). Granting the relation between u(t, x) and V(t, X), we prove tha t the inequality holds true for smooth functions u(t, x) satisfying the condition (5):

The validity of the inequality (14) for any proved by passing to the limit. L e m m a 8 . For any function

can be the following


P r o o f . At first, let us prove the inequality for smooth functions V(t, X) satisfying the conditions (16). Introduce an auxiliary functio n u(t, x) in the following way:

The definition of the function u(t, x)

Note that

implies that


223

Let us prove that the following inequality holds true Consider

Using the operation of differentiation by parts, relation between u(t, X) and v(t, x ), and initial conditions, we obtain

Granting initial and boundary conditions, we have

224

In a similar way,

Chapter 7


225

226

It is follows from the fact that the matrix and nonnegative that

Next,

Chapter 7

is symmetric


227

228

Chapter 7

Note that it is possible to prove the convergence of the improper integrals, which are the terms of the sums. Thus,

The relations mentioned above implies the fact that (17) holds true. Applying the Schwartz inequality to (17), we obtain:


Reducing by

229

and accounting the relation between v(t, x )

and u(t, x) , we obtain that the inequality holds true for smooth functions v(t, x) satisfying the conditions (7):

The validity of the inequality (16) for any function proved by passing to the limit. T h e o r e m 1. For any function

is

there exists

a unique generalized solution of the problem (12) in the sense of Definition 1. The similar statement holds true for the adjoint problem. P r o o f . Theorem 1 follows from general Theorem 1.1.1 Next, we shall investigate the solvability of the problem

where u(t, x ), v(t, x ) are unknown elements, and f, g are given elements. By the solutions of the problems (18), (19) we shall mean the generalized solutions in the following sense.

D e f i n i t i o n 3. The generalized solution of the problem (18) is a function such that there exists a sequence of smooth functions

which satisfy the condition (5) and

230

Chapter 7

It should been stressed that the difference between generalized solution in the sense of Definition 1 and the generalized solution in the sense of Definition 3 is that the corresponding sequences of smooth functions must converge in the different metrics. D e f i n i t i o n 4 . The generalized solution of the problem (18) is a function such that the identity

holds true for any smooth function v(t, x)

satisfying

the

conditions (7). In a similar way, we can define a generalized solutiion of the adjoint problem. T h e o r e m 2. For any function

there exists

a unique generalized solution of the problem (18) in the sense of Definition 3. Analogous statement holds true for the adjoint problem. R e m a r k . The inequalities (14), (16) are not only sufficient but also necessary conditions of the existence of a unique generalized solution. Corollary. The inequalities (14), (16) imply more positive inequalities:

P r o b l e m 5 . We shall investigate the solvability of the following problems


where u(t, x), v(t,x) are unknown elements, and

231

f,

g are

given elements. By solution of problem (22), (23) we shall mean a generalized solution in the following sense. D e f i n i t i o n 5. The generalized solution of the problem (21) is a function such that there exists a sequence of smooth functions

satisfying the conditions (5) and

The generalized solution of the adjoint problem is defined in a similar way. L e m m a 9 . For any functions and the following inequalities hold true

The proof of Lemma 9 is similar to the proofs of Lemmas 7 and 8. T h e o r e m 3. For any function there exists a unique generalized solution of the problem (22) in the sense of Definition 5. Analogous statement holds true for the adjoint problem. P r o o f of Theorem 3 is analogous to the proof of Theorem 1. Next, we shall investigate the solvability of the following problems

Chapter 7

232

where u(t, x ), v(t,x) are unknown elements, and f , g are given elements. By solutions of the problems (26), (27) we shall mean generalized solutions in the following sense. D e f i n i t i o n 6. The generalized solution of the problem (26) is a function of smooth fucntions

such that there exists a sequence satisfying the conditions (5) and

In a similar way, the generalized solution of the adjoint problem is defined. It should been stressed that the difference between the generalized solution in the sense of Definition 5 and the generalized solution in the sense of Definition 6 is that the corresponding sequences of smooth functions must converge in the different metrics. T h e o r e m 4 . For any function there exists a unique generalized solution of the problem (26) in the sense of Definition 6. The analogous statement holds true for the adjoint problem. Remark. The inequalities (24), (25) are not only sufficient, but also necessary conditions of the existence of a unique generalized solution. Corollary. From the inequalities (24), (25) follows more positive inequalities


233

P r o b l e m 6 . We shall investigate the solvability of the following problems

where u(t, x), v(t, x) are unknown elements, and f , g are given elements. By solution of the problems (30), (31) we shall mean a generalized solution in the following sense. D e f i n i t i o n 7. The generalized solution of the problem (30) is a function

such that there exists a sequence

of smooth functions satisfying the conditions (5) and

The generalized solution for the adjoint problem is defined in a similar way. L e m m a 10 . For any functions

the following


P r o o f . Let us prove the inequality for smooth functions u(t, x) satisfying the conditions (5). Introduce an auxiliary operator in the following way:

234

Chapter 7

where

If the function u(t, x) satisfies the boundary conditions (5), then the function v(t, x) defined above satisfies (7), furthermore, the following relation holds true:

Denote the expression

Applying the operation of integration by parts, the OstrogradskyGauss formula, and granting the boundary conditions in the passing to the integration on the boundary, we obtain

where


235

236

Chapter 7

Therefore,

Applying the Schwartz inequality ,to the left-hand side of (34), we have:

Hence,

From the relation (33) we obtain

Thus, the inequality (32) holds true for smooth functions u( t, x) satisfying the conditions (5). The validity of (32) for any is proved by passing to the limit.


L e m m a 1 1 . For

any

237

functions

the

following inequality holds true

P r o o f . The proof of Lemma 11 is similar to the proof of Lemma 10. The integral operator in this case has the following form

where

T h e o r e m 5. For any function

there exists a

unique generalized solution of the problem (30) in the sense of Definition 7. The analogous statement holds true for the adjoint problem. The proof of Theorem 5 is similar to the proof of Theorem 1. Next, we shall investigate the solvability of the following problems

where u(t, x), v(t, x) are unknown elements, and f , g are given elements. By solution of problems (36), (37) we shall mean a generalized solution in the following sense.

238

Chapter 7

D e f i n i t i o n 8 . The generalized solution of the problem (36) is a function such that there exists a sequence of smooth functions

satisfying the conditions

(5), and

The definition for the adjoint problem is similar. It should been stressed that the difference between the generalized solution in the sense of Definition 7 and the generalized solution in the sense of Definition 8 is that the corresponding sequences of smooth functions must converge in the different metrics. T h e o r e m 6. For any function

there exists a

unique generalized solution of the problem (36) in the sense of Definition 8. The analogous statement holds true for the adjoint problem. The proof of Theorem 6 is similar to the proof of Theorem 2. R e m a r k . The inequalities (32), (35) are not only sufficient, but also necessary conditions of the existence of a unique generalized solution. Corollary. The inequalities (32), (35) implies the positive inequalities (obtained by many authors):


239

2. GENERALIZED SOLVABILITY OF PSEUDOHYPERBOLIC SYSTEMS (THE NEUMANN INITIAL BOUNDARY VALUE PROBLEM)

2.1 Main notations and auxiliary statements Consider the linear partial differential equation:

in a tube domain

where u(t,x) is an unknown

function depending on spatial variable is a bounded simply connected domain in The operators

and

and time with smooth boundary

do not depends on t and they are

defined by the following differential expressions:

and

are

Functions defined

in

a

closed

domain Let be continuously differentiable, and

C(x) , D(x) be continuous in the closed domain

functions.

240

Chapter 7

We shall suppose that the differential expression (2) is uniformly elliptic, and the expression (3) is non-negative in i.e. for any and any the following relations hold true

where are constants which do not depend on X and Suppose also that D e fi n i t i o n 1 . (denoted as

the gradient of the function u(t, x) is a vector-column

where

In the matrix form where A and B are matrices n × n consisting of the elements


D e fi n i t i o n 2 . with respect to the normal

241

of the function u(t, x) to the surface ( denoted as

is the following inner product in

In the expanded form

where

is i -th

can be written as

component of the vector

at the point

of the surface Let us introduce the following notations: (Q) is the space of measurable square integrable on the set functions, D(L) is the set of smooth (infinitely differentiable) in following conditions:

functions, which satisfy the

We shall assume that the operator maps (Q) into (Q) and has the domain of definition Note that the set D(L) is dense in the space (Q), and therefore it is possible to define correctly the adjoint operator whose domain of definition is the set of functions v(t, x) for which there exists an element (such element is equal to following identity holds true for any Thus, the formally adjoint operator has the following form:

that the

242

Chapter 7

with boundary conditions

The notations introduced above give the possibility to consider the second initial boundary value problem. P r o b l e m 1 . To find a function u(t, X), which satisfies the equation (1) in the domain and the conditions (4) on the boundary P r o b l e m 2 . To find a function v(t,x ) which satisfies the equation (5) in the domain and the conditions (6) on the boundary Note that simultaneously with the common equation of pseudohyperbolic type (1) we shall investigate also some simpler version of this equation, for which it is possible to obtain more interesting results. Let

where the operator and the functions C(x) , D(x) satisfy the same conditions as in the case of the operator and be a nonnegative constant. Let the domain of definition of the operator is the set conditions

where

of smooth in

functions u(t, x) , which satisfy

is a co-normal to the surface

which is defined as:


243

where is an outward normal to the surface n × n consisting of the elements

A is a matrix

Note that despite of the difference between the boundary conditions (4) and (8) all results, which shall been obtained for the operator can be easily adopted for the operator We shall consider the operator

where In this case the domain of definition of the operator is the set D(L) . Thus, the operator converts to the operator if in we put Since, all results concerning the operator are valid for the equation (9). But in contrast to the operators and which map the operator can be considered in other pairs of spaces also. Denote by the completion of the set of smooth in functions with respect to the norm generated by the following inner product:

Since the boundary conditions (8) are not held in this norm (10), it is clear that the set is dense in the space as far as by completing the set space

in the norm (10) we shall obtain the same

(it is the same situation as in the case of density of finite

functions of the class

in the space

(Q) ). Taking into account

244

the density of maps adjoint operator

Chapter 7

in

we shall consider that the operator and also we shall define the formally

with the conditions

Now it is possible to formulate the following problems. P r o b l e m 3 . To find a function u(t, x) satisfying the equation (9) in the domain Q and the conditions (8) on the boundary P r o b l e m 4 . To find a function v(t, x) satisfying the equation (11) in the domain Q and the conditions (12) on the boundary Since the right-hand side of Problems 1-4 can be a discontinuous function or even a Schwarz distribution then it is possible that there are no any classic solutions of this equations, therefore we must consider some generalizations of the solutions and the extensions of the operators and respectively. Let us introduce the following notations. Let completions of the set D(L) in the norms

be


respectively;

245

be completions of the set of smooth in

functions satisfying the conditions (6) in the same norms (13)-(15), respectively. It is easy to test whether the expressions (13)-(15) satisfy the norm axioms using the boundary conditions (4), (6) and Minkowsky inequality. Let us build on the pairs

and

and the negative spaces respectively, as completions of the set

and also and in the norm

and

respectively. In a similar way we can build the negative spaces and

on the pairs

and

and also

and

The following imbeddings are valid:

Let us introduce some additional notations. Let completions of the set

in the norms

be

246

Chapter 7

respectively;

be completions of the set of smooth in

functions satisfying the conditions (12) in the norms (20)-(22). Let us introduce the negative spaces this case we construct their on the pairs also

and

and and

as completions of the set

but in and

in the norms

and

In a similar way we can construct the spaces Now, the following imbeddings are valid:

and


247

2.2 A priori inequalities in negative norms L e m m a 1 . For arbitrary smooth in functions u(t,x) and v(t, x) satisfying the conditions (4) and (6), respectively, the following inequalities hold true: where the constant c does not depend on the functions u(t, x) and v(t, x). P r o o f . By definition of the negative norm, we have

Consider the numerator. Applying to it the formula of integration by parts, the Ostrogradsky-Gauss formula and taking into account the conditions (4) and (6), we obtain

Applying to the right-hand side the Cauchy inequality ,we have

248

Taking into account that

Chapter 7

C(x) , D(x)

are

continuous in and since, they are bounded, and granting the inequalities (16) and (21), it is easy to see that every term in the righthand side does not exceed and as far as the number of these terms is finite then This completes the proof of the theorem. The proof of the second inequality is the same. In a similar way it is possible to prove the following three lemmas. L e m m a 2 . For arbitrary smooth in functions u(t, x) and v(t, x) satisfying the conditions (4) and (6), respectively, the following inequalities hold true: where a constant c does not depend on the function u(t, x ) and v(t, x). L e m m a 3 . For arbitrary smooth in functions u(t, x) and v(t, x) satisfying the conditions (8) and (12), respectively, the following inequalities hold true: where a constant c does not depend of the functions u(t, x) and v(t, x). L e m m a 4 . For arbitrary smooth in functions u(t, x) and v(t, x) satisfying conditions (8) and (12), respectively, the following inequalities hold true where constant c does not depend on the functions v(t, x).

u(t, x) and


249

Note that Lemmas 1-4 give the possibility to consider some extensions of the operators

and

For example, the first inequality of Lemma 1 gives the possibility to extend with respect to continuity the operator and consider it as an operator mapping

into

We save the

previous notations for the extended operators and as far as they will be specified by context. Note that the inequalities in Lemmas 1-4 hold true for the extended operators also but on the whole space The proof of this fact follows from the passing to limit in the inequalities for smooth functions. L e m m a 5 . Let u(t, x) be an arbitrary smooth in function satisfying the condition (4), and operator defined by the following expression:

where

Then,

is the integral

250

P r o o f . At first, note that

Chapter 7

Indeed, the initial

conditions in (6)

are valid at the expense of the form of the operator the boundary conditions in (6)

Concerning

it should been noted that the function not necessarily satisfies this condition, as far as in completion of the set of smooth functions in the norm (13) this condition becomes not valid. Let us express u(t, x) through v(t, x) using (13).

Consider the left-hand side (32). Applying the formula of integration by parts, the Ostrogradsky-Gauss formula and the conditions (4), (33), which are satisfied by the functions u(t, x) and v (t, x) we have

Consider every of the last terms separately. l. Then,

Applying the formula of integration by parts to the second term, we obtain


251

Whence using the Ostrogradsky-Gauss formula and the initial conditions, we obtain

Thus,

Granting the conditions imposed on

and

we have

2. Let us apply the formula of integration by parts to the second term in (34):

Using the Ostrogradsky-Gauss formula, we obtain

As far as u(t, x) satisfies the conditions (4), then the first integral equals to zero. Let us execute some transformations with the second

252

Chapter 7

integral. We shall apply to it the formula of integration by parts and the Ostrogradsky-Gauss formula:

Since u(0,x)= 0 and v (T, x) = 0 then the first integral equals to zero, and therefore

Consider every of the last integrals separately. a) Granting that we have

Applying to the first integral the formula of integration by parts, we obtain:

Passing in the first term to the integral on surface and taking into account the conditions of the uniform ellipticity of and also the relations

we have


253

b) Consider the second term in (35):

Applying to the first integral the formula of integration by parts, we have:

In the first term we pass to the surface integral, join the second term with the fourth, apply the formula of integration by parts, and take into account the condition for the third term

Then we obtain

254

Taking into account the fact that the operator and the values of the constants and we obtain

Chapter 7

is non-negative

Thus, taking into account the results of the subsections a) and b), and returning to the identity (35), we obtain


255

Convince us that the second and the third terms in the right-hand side are non-negative. Indeed, as far as

then

Therefore, the second term is non-negative. Furthermore,

that proves the non-negativity of the third term. Finally, we have

Consider the third and the fourth terms in (34). 3. For the third term we have the following estimation:

256

Chapter 7

Applying to the integral in the right-hand side the formula of the integration by parts and the Ostrogradsky-Gauss formula, we obtain

Thus, 4. The fourth term is estimated in the following way:

Integrate by parts the first term in the right-hand side and pass to the surface integral:

Apply again the formula of integration by parts to the last term:


257

Hence,

Thus, every of the four terms in (34) are considered. Taking into account the transforms carried out, we have

This proves the lemma.. Corollary. For an arbitrary function

the

following inequality holds true: P r o o f . The lemma implies that for an arbitrary smooth function u(t, x) satisfying the condition (4) the following inequality holds true:

Apply to the left-hand side the Schwartz inequality:

258

Reducing by

Chapter 7

we obtain the inequality

Now we must to justify the following inequality:

that proves the required inequality for smooth functions u(t, x) . Taking into account that operator is continuous and passing to the limit, we prove the required inequality for an arbitrary function L e m m a 6 . Let v(t, x) is an arbitrary smooth in function satisfying the conditions (6), and operator defined by the following expression:

is the integral

where

Then, The proof of this lemma is similar to the previous one, therefore we shall write only calculations. The inverse operator to

is


Applying to

259

the formula of integration by parts, we

have

Consider every of the four terms of the right-side hand separately . 1. The first term is estimated in the following way:

2. Now, let us estimate the second term.

260

Integrating by parts, we obtain

Taking into account the relations

and integrating by parts, we have

Chapter 7


Making obvious reductions, we obtain

3. The third term is estimated in the following way:

261

262

Chapter 7

Integrating by parts, we have

Thus,

Corollary. For an arbitrary function

the

following inequality holds true The proof is similar to the proof of the corollary of Lemma 5. Thus, granting Lemma 1 and the corollaries of Lemmas 5 and 6, we have


263

D e f i n i t i o n 3 . A generalized solution of Problem 1 is such function that there exists a sequence of smooth in

functions


D e f i n i t i o n 4. A generalized solution of Problem 1 is such function that there exists a sequence of smooth in

functions


L e m m a 7. Let u(t,x ) be an arbitrary smooth in function satisfying the conditions (4), and operator defined by the following expression:

is the integral

Then,

P r o o f. As in the proof of Lemma 5 note that Indeed, the initial condition v(T, x) = 0 holds true at the expense of the form of the operator Remaining conditions (6) do not hold true after completion of smooth functions in the norm of the space Consider where

264

Chapter 7

Consider every of these term separately. 1. Applying the formula of integration by parts, we obtain

To calculate the first integral we use the Ostrogradsky-Gauss formula and also the conditions u(0, x) = v(T, x) = 0 . In the second integral we take into account that (37) implies that

Integrating the last expression by parts, we obtain

2. Applying the formula of integration by parts, the OstrogradskyGauss formula and taking into account the condition we have


265

Consider both terms separately. a) To calculate the first integral, we use the formula of integration by parts. We obtain that

Passing to the integral on surface in the first term and taking into account the conditions u(0, x) = v(T, x) = 0, we have that this integral equals to zero. To calculate the second integral, we use the condition and also take into account the uniform ellipticity of the operator

b) Consider the second term. It is clear that the following identity holds true

266

Chapter 7

Let us prove that the last expression in non-negative. To do this, we apply the operation of integration by parts and take into account the conditions

Hence,

Thus,

Consider and 3. Similarly to the previous reasoning we have

4. Analogously,


267

Thus, summarizing all written above, we have

Let us prove that

Indeed,

Hence,

The lemma is proved.

Corollary. For an arbitrary function following inequality holds true

the

P r o o f . Applying to the left-hand side of the inequality the Schwartz inequality, we have:

Reducing by

we obtain

Taking into account (38), we obtain

that proves the required inequality for smooth functions u(t, x) . Taking into account that the operator is continuous and passing to the limit we find that the required inequality holds true.

268

Chapter 7

L e m m a 8. Let v(t, x) be an arbitrary smooth function satisfying the conditions (6), and operator defined by the following expression

is the integral

Then, The proof of this lemma is similar to the previous, therefore we represent it in brief version. Let us write where

Thus,


269

Taking into account the form of the operator the required inequality. Corollary. For an arbitrary function following inequality holds true

we obtain the

The proof of the lemma is similar to the proof of Lemma 7. Thus, by Lemma 2 and Corollaries of Lemmas 7 and 8, the following inequalities hold true:

D e f i n i t i o n 5. A generalized solution of Problem 1 is such function that there exists a sequence of smooth in

functions

(t, x) satisfying the conditions (4) and

D e f i n i t i o n 6 . A generalized solution of Problem 1 is such function that there exists a sequence of smooth in functions satisfying the conditions (4) and In a similar way the definitions for Problem 2 are formulated. L e m m a 9. Let u(t, x)is an arbitrary smooth function satisfying the conditions (8), and defined by the following expression

where

is the integral operator

270

Chapter 7

Then,

P r o o f. It is easily to see that

as far as all the

conditions (12) hold true. Consider where

is the inverse operator to

and the operator:

Then,

The first term is considered in a similar way as in Lemma 5:

where


Consider the second term in (40)

where

Integrating the second term by parts, we have

271

272

Chapter 7

Note that granting that

and the conditions (8), we have

Therefore, applying to the first integral the Ostrogradsky-Gauss formula, we obtain that it equals to zero. Thus,

Similarly to the previous case, we have

Applying to the first integral the operation of integration by parts, we obtain


Let us integrate by parts the second term again:

Integrating by parts the first term, we obtain:

Whence, we have

273

274

Thus, granting that the operator

Chapter 7

is uniformly elliptic, we finally

obtain:

Taking into account the values of the constants out the obvious estimations, we obtain

and carrying


275

Taking into account the uniform ellipticity of the operator obtain

we

Note that for the elliptic operators the inequality of coercitivity holds true [74]:

Taking into account the inequality of coercitivity with s = 0, we have

Corollary. For an inequality holds true

arbitrary

the following

The proof of is similar to the previous proofs of the corollaries of the lemmas.

276

Chapter 7

L e m m a 1 0 . Let v(t, x) is an arbitrary smooth in function from

and

v is the integral operator

where

the constants

and

are defined as

and

Then, P r o o f . Let us express from v(t, x) through u(t, x) from (41). Consider

where


277

278

Chapter 7


Finally, we have

279

280

Chapter 7

Applying the inequality of coercitivity as in Lemma 9, we obtain that the lemma holds true. Corollary. For an arbitrary function the following inequality holds true:

The proof is similar to the previous ones. Thus, by Lemma 3 and the corollaries of Lemmas 9 and 10, we find that the following inequalities hold true

for an arbitrary functions

and

D e f i n i t i o n 7. A generalized solution of Problem 3 is such function that there exists a sequence of smooth in functions satisfying the conditions (8) and


281

D e f i n i t i o n 8 . A generalized solution of Problem 3 is such function that there exists a sequence of smooth in

functions


In a similar way the definitions for the adjoint Problem 4 are introduced. L e m m a 11 . Let u(t, x) be an arbitrary smooth in function satisfying the conditions (8), and operator defined by the following expression:

is the integral

Then

The proof is similar to the previous ones. C o r o l l a r y . For an arbitrary function following inequality holds true

the

The proof is similar to the previous ones. L e m m a 12 . Let v(t, x) be an arbitrary smooth in function satisfying the conditions (12), and integral operator defined by the following expression

Then

is the

282

Chapter 7

The proof is similar to the previous Lemma. C o r o l l a r y . For an arbitrary function following inequality holds true

the

By Lemma 4 and the corollaries of Lemmas 11 and 12 the following inequalities hold true

D e f i n i t i o n 9 . A generalized solution of Problem 3 is such function that there exists a sequence of smooth in functions satisfying the conditions (8) and D e f i n i t i o n 1 0 . A generalized solution of Problem 3 is such function that there exists a sequence of smooth in

functions


The definitions of Problem 4 are formulated in a similar way. Using the results obtained above and applying the general theorems from Chapter 1, we obtain the following there T h e o r e m 1 . For an arbitrary functions exists a unique solution u(t, x) of Problem 1 in the sense of Definition 3. T h e o r e m 2 . For an arbitrary function there exists a unique solution u(t, x) of Problem 1 in the sense of Definition 4.


283

T h e o r e m 3 . For an arbitrary function , respectively) there exists a unique solution u(t, x) of Problem 1 in the sense of Definition 5 (Definition 6, respectively). T h e o r e m 4 . For an arbitrary function respectively) there exists a unique solution u(t, x) of Problem 3 in the sense of Definition 7 (Definition 8, respectively). T h e o r e m 5 . For an arbitrary function respectively ) there exists a unique solution u(t, x) of Problem 3 in the sense of Definition 9 (Definition 10, respectively). The proofs of these theorems follow from general Theorems 1.1.1 and 1.1.3 Similar theorems hold true for Problem 2 and 4, also.

3. ANALOGIES OF GALERKIN METHOD FOR PSEUDO-HYPERBOLIC SYSTEMS Consider Galerkin’s method for second boundary value problems for pseudo-hyperbolic systems, which was studied in paragraph 2 of the chapter. All denotations of the following correspond to paragraph 2 ones. Let the right-hand side f(t, x) of equation (2.1) be a smooth in function, that satisfies the condition: Consider approximate solution of problem (2.1) in the following form where is a solution of the Cauchy problem for the set of linear ordinary differential equations with constant coefficients.

284

where

Chapter 7

is a sequence of smooth in is total in

functions, that the set is a smooth function:

Denote by H' the completion of set of smooth in satisfy the conditions (2.4), in the norm

functions that

L e m m a 2 . Consider the differential operator

then for an arbitrary function inequality is true

where by the bilinear form

u(t, x)

we mean

the

following


285

Denote by H' the completion of set of smooth in satisfy the conditions (2.4), in the norm

functions that

L e m m a 2 . Consider the differential operator

then for an arbitrary inequality is true where by the bilinear form

function

the following

we mean

P r o o f . Consider an arbitrary smooth in function u(t, x), that satisfies the conditions (2.4). Applying he Schwarz inequality, we have

Taking into C(x), D(x),

account

the continuity in the domain

of the functions and a(t),b(t) in

[0, T], we have that each norm in the right-hand side not exceed the that is what had to be proved for the right-hand side of the

286

Chapter 7

desired inequality. To prove the left-hand side, consider the following expression where u is a smooth function. Transforming the

in the much the same way as in Lemma 2,6, we have

the desired inequality for smooth functions. To prove the inequality for all functions u form H' it is necessary to pass to the limit. Let u(t, x) be a function from H' and be a sequence of smooth functions, that satisfy the conditions (2.4) and Since

then the sequences converge to

respectively, in the space

(Q) (by the derivatives we mean derivatives of the distributions). It is clear, that

thus

It is what had to be proved. L e m m a 3. Consider the

differential

then for an arbitrary function following inequality is true

operator the


P r o o f . Let u(t, x) be a smooth in conditions (2.4). Consider

287

function, that satisfies the

where

Thus,

The verity of the inequality for all functions we obtain by passing to the limit. The right-hand side of the inequality is proved in the same way as the previous one. L e m m a 4 . For the functions which have been defined in (1), the following inequality is true

288

Proof.

Consider

where

Thus,

Chapter 7

Let

Prove

that


Prove, that hand sides of (2) by

289

Multiplying the both right and left and summing up over j from 1 to s, we

obtain

Substitute t = 0 into the relation and take into account that Thus,

Consider the relation (2) again. Differentiating the relation (2) with respect to t, multiplying by

summing up over j from 1 to

s and integrating with respect to t from 0 to T, we have

From the previous inequality we find

It remains to mark that proved.

which is what had to be

290

Chapter 7

By Lemma 4, we obtain that the sequence the space H', thus there exists a function

is bounded in and a subsequence

weakly in H'. It is clear that the sequences of norm

are bounded (since

is bounded in H' ), and then there exists

subsequence (which is denoted by

weakly in that

again), that

Since

weakly, it is easy to prove

where derivatives of the function are understood in the sense of distributions. T h e o r e m 2. For all smooth in the domain function the approximate sequence

converges to

the solution of Problem 2.1 in the sense of Definition 2.3 in the norm of space P r o o f. Multiplying the relation (2) by


291

summing up over j from 1 to s and integrating with respect to t from 0 to T, we obtain

thus,

where the sequence

converges weakly to

in the space

H' . Taking into account the relation (4), we have that so

From Lemma 2 we find

Prove that Multiplying

the

equality

(2)

by

a

function

summing up over j form 1 to p and integrating with respect to t from 0 to T, we have

292

Chapter 7

Let

then

From the equalities (4), we obtain Therefore,

By virtue of the totality of the system there exists sequence

That is why

in the space in space

then

Returning to (6), we

have

From (4), we find

Passing in the (7) to the limit and taking into account (5)

Therefore, there exists a function which converges to the function

and a sequence in the norm of space

Prove


293

that the function is a solution of Problem 2.1. Applying the inequality of Lemma 2.1, we have

So that, the sequence


thus there exists an element Prove that function

in the space

Multiply the equality (2) by a

h(t) : h(T) = h ' ( T ) = 0 , and

let

Integrating the both right and left-hand sides of the equality (8) with respect to t from 0 to T, we have: Pass to the limit as

Since the totality of the set

in the space

Using the convergence of the sequence

we obtain and (8), it’s

easy to prove that the function is a solution of Problem 2.1 with the right-hand side f ( t , x) in the sense of Definition 2.3. It remains to mark that by Theorem 2.1 the solution is unique, thus it is not necessary to choose the subsequence accumulation point of the sequence

If there is an in fact that differ from

, then by the same reasoning Problem 2.1 has another solution. T h e o r e m 3 . For an arbitrary smooth in function the approximate sequence

converges to

294

Chapter 7

the solution of Problems 2.1 in the sense of Definition 2.5 in the norm of space and P r o o f is analogous to the reasoning of previous theorem. Now, consider the case Consider the approximate sequence in the form of the relations (l)-(3). L e m m a 5 . The following inequality is valid P r o o f . Reasoning as in the proof of Lemma 2.6, we prove that for a smooth in function u(t, x), which satisfy conditions the following inequality

is

valid. Choose the sequence of smooth on [0,T] functions that converges to the solution of the set of the differential equations in the space where Sobolev space of functions v(t) : v(0) = v´(0) = 0 . Prove that

where Actually,

is the


295

Therefore, Since the function Lemma 2 is valid. Thus,

It is easy to show that

is smooth in We have

then

296

Chapter 7

thus passing to the limit in the inequality (9) as

we obtain that

Now, we return to the relations (2). Multiplying the relations (2) by

summing up over j from 1 to s and integrating with respect to t from 0 to T, we obtain Whence,

To

prove

the

lemma,

it

suffices

to

mark

that

The lemma is proved. T h e o r e m 4. For an arbitrary function the approximate sequence

converges to the solution of

Problem 2.1 in the sense of Definition 2.3 in the norm of space and P r o o f . Choose the sequence of smooth in

such that

functions

Let

be the

approximate sequence (1) of the problem with the right-hand side By Theorem 2, the sequence Prove that the sequence In fact, by Lemma 5 we have

converges to the solution

as

is fundamental in the space


297

By Theorem 2,

and From the other hand, since

the sequence

is fundamental, whence there exists such element

Therefore, that

Consider By Lemma 4 Thus, Approaching to the limit as Making

we have

we have

Prove that is a solution of Problem 2.1 with the right-hand side f(t, x) in the sense of Definition 2.3:

298

Chapter 7

Prove that every summand in the right-hand side vanishes. By Lemmas 2.6 and 5, we have

By Theorem 2,

Thus,

Applying (10) and (11), it is easy to prove that is a solution of Problem 2.1 by Definition 2.3. Now the theorem is proved. Let and the approximate sequence of solution is defined in (l)-(3) (in the case the integrals are defined in the sense of distribution theory). L e m m a 6 . The following inequalities is valid The lemma can be proved in much the same way as Lemma 5 (it is necessary to substitute the space for the space the operator for and to apply the Schwarz inequality to the right-hand side of the following expression .

T h e o r e m 5. For every function sequence

the approximate

converges to the solution of Problem 2.1 in the

sense of Definition 2.5 in the norm of the space T h e P r o o f is completely analogous to that of Theorem 4.

and


299

Now, let f(t, x) be an element of the negative space a

sequence

of

functions

from

Consider the approximate sequence

where functions

be such

that

of the form

are solutions of the following system

T h e o r e m 6 . Let

be an arbitrary number sequence Therefore, for each integer s(m) : (which exists necessarily), and for each

right-hand

side

the approximate

sequence

converges to the solution of Problem 2.1 in the sense of Definition 2.6 in the norm of space

300

Chapter 7

P r o o f . By Theorem 4 the approximate sequence converges to the solution of the following equation then there exists such function

Prove that the sequence

with

that

is fundamental in the space

Using Lemma 2.5, we have

Pass to the limit as

and

there

exists

Then,

a

function

Consider

Choose

s = s(m) such

that

Theorem 4, the number s = s(m) exists). Thus,

(by


301

and

which is what had to be proved. If f (t, X) be an element of the negative space an

arbitrary

sequence

of

then choose

functions

Approximate sequence

(t,x) is defined

in by the relations (12)-(14). In this case the following analogous of Theorem 6 can be proved. T h e o r e m 7. Let be an arbitrary number sequence Therefore,

for

each

integer s(m) :

(which exists necessarily), and for each right-hand side the approximate sequence converges to the solution of Problem 2.1 in the sense of Definition 2.6 in the norm of space Let the right-hand side f (t,x) of the equation (2.1) is a smooth in function, that satisfies the following condition Approximate solution can be found in the form

where the function (t) is a solution of the Cauchy problem for the set of differential equations with constant coefficients:

302

Chapter 7

where

where the condition where

is a sequence of smooth in and the set

functions, which satisfy is total in

is an arbitrary smooth function:

Denote by the completion of the set of smooth in functions, which satisfy the condition (2.4), in the norm


L e m m a 7. Let operator:

be

then for every function valid

where by

303

the

following

differential

the following inequality is

we mean

P r o o f . Let u(t,x) be a smooth in function, which satisfy the condition (2.8). Then, the right-hand side of the required inequality we obtain by applying partial integration. To prove the left-hand side, it is necessary to repeat the transformation of Lemma 2.10 with the expression

The required inequality in the space

can

be obtained by passing to the limit.

L e m m a 8 . Let

be

the

following

operator: then for every function following inequality is valid

differential the

P r o o f is analogous to that of previous lemma. L e m m a 9 . For functions which were defined in (15)-(17), the inequality is true

P r o o f . Prove that

304

Chapter 7

Consider

where

Transform each summand as

and

in Lemma 4. Hence,


305

Therefore, we have

In a way analogous to Lemma 4, we prove that

Taking into account the inequality of coercivity, we obtain

Differentiating the equality (16) with respect to , multiplying on summing up over respect to

from 1 to s and integrating with

from 0 to T, we have

306

Chapter 7

Whence, we find

It suffices to remark that T h e o r e m 8 . For

all

smooth

in

the approximate sequence

functions (see (15)-(17))

converges to the solution of Problem 2.3 in the sense of Definition 2.7 in the norm of space and T h e o r e m 9 . For

all

smooth

the approximate sequence

in

functions converges to

the solution of Problem 2.3 in the sense of Definition 2.9 in the norm of space and Let f (t, x) be an arbitrary function from sequence

and approximate

is defined by the equalities (15)-(17). Using the

results of Lemma 2.10 and repeating the proof of Lemma 5, the following lemma can be proved. L e m m a 1 0 . The following inequality is true T h e o r e m 1 0 . For approximate sequence

all

functions

the


Problem 2.3 in the sense of Definition 2.7 in the norm of space and


307

In the case when the function f (t,x) is from the space the analogous results of convergence of the approximate solution (15)-(17) can be obtained. L e m m a 11 . The following inequality is true T h e o r e m 11 . For approximate sequence

all

function

the


Problem 2.3 in the sense of Definition 2.9 in the norm of the space and

Let f (t,x) be an arbitrary function from of functions

from

such that

Choose a sequence and

consider an approximate sequence of the solution •

where (t) is a solution of the Cauchy problem, for the set of linear differential equations with constant coefficients

where the constants

were defined in (18), the sequence

satisfies the same conditions as in (15)-(18). T h e o r e m 1 2 . Let be an arbitrary number sequence Therefore, for each integer s(m) :

308

Chapter 7

(which exists necessarily), and for each right-hand

side

the approximate

sequence

converges to the solution of Problem 2.3 in the sense of Definition 2.8 in the norm of space Analogously for

we have.

T h e o r e m 1 3 . Let

be an arbitrary number sequence

Therefore,

for

each

integer

s(m) :

(which exists necessarily), and for every right-hand

side

the approximate

sequence

converges to the solution of Problem 2.3 in the sense of Definition 2.10 in the norm of space

4. PULSE OPTIMAL CONTROL OF PSEUDOHYPERBOLIC SYSTEMS (THE DIRICHLET INITIAL BOUNDARY VALUE PROBLEM) Apply the obtained results to optimization of the pseudo-hyperbolic systems. The denotations are the same as previous. Consider the pseudo-hyperbolic equation

where is a one of the pseudo-hyperbolic operator (see Section 1). Using the template theorems od Section 3.6, complete the tables for the pseudo-hyperbolic equation (the first boundary value problem).


309

310

Chapter 7


311

312

Chapter 7

5. PULSE OPTIMAL CONTROL OF PSEUDOHYPERBOLIC SYSTEMS (THE NEUMANN INITIAL BOUNDARY VALUE PROBLEM) Apply the results of Section 3.6 to optimization of the pseudohyperbolic systems (the Neumann initial boundary value problem). The denotations are the same as in Section 2 of this chapter. Consider the pseudo-hyperbolic equation where is a one of the pseudo-hyperbolic operator Section 1) with corresponding boundary conditions:

(see

Using the template theorems of Section 3.6, complete the tables for the pseudo-hyperbolic equation (the Neumann initial boundary value problem).


313

314

Chapter 7


315

Remark that if we consider the problem (1), (2) with operator other theorems can be proved, because of the theorems of the solvability of the equation ensure smoother solution when the right-hand side belongs to some positive spaces. Let be a negative space corresponding to the pair of spaces (Q) . Consider the problem of optimal control of the system (1), (2) with the right-hand side

It requires to minimize the

performance criterion

where

is functions from

T h e o r e m 1 0 . Let the state function problem

(1),

(2)

with

the

and

and

u(t, x) satisfy the

right-hand side

316

Chapter 7

Performance criterion is in the form (4). If there exists a Fréchet derivative of the map then the performance criterion J(h) has a Fréchet derivative at the point following form

where

is a bilinear form on

in the

and v(t, x) is a

solution of the adjoint problem'

P r o o f.

Prove, that

Theorem 2.3 the solution and

Since belongs to

by

Since the spaces

have the equivalent norms, then we can consider that Applying the analogue of Theorem 2.4 for

the

adjoint

operator,

we

have

Thus

is a linear continuous functional on H. From the analogue of Theorem 2.4 we have that for all functions the following equality is true

In the same way as Lemmas 2.1-2.4, we can prove that for all functions


thus

317

Therefore the equality (8) is valid for all

Taking account that

we have

In a way analogous to that was made in Theorem 1.4,3, we obtain

where satisfies the equation

It is clear that the function

where By Theorem 2.3, we have that

Since

and

we have that and

Consider the equation

As is well known from the theory of elliptic equations, the solution of the equation exists and has the degree of the smoothness with

318

Chapter 7

respect to the space argument two orders higher than the right-hand side of the equation Thus,

and it is easy to see that

Therefore,

Substituting

Since

into (9) and taking into account (11), we have

we obtain

Using the condition

we have


319

Returning to (10), we have that

The further part of the proof repeats the reasoning of Theorem 2.1.3. T h e o r e m 11 . If the mapping has a continuous

Fréchet

derivative

at

a

point then

the derivative is continuous at the point P r o o f is analogous to that of Theorem 2.1.4. T h e o r e m 12 . If mapping

has a Fréchet

derivative in a neighbourhood of a point Lipschitz condition with exponent

that satisfies

from

point

the

neighbourhood

of

the

then the Fréchet derivative satisfies Lipschitz condition with the same exponent P r o o f is analogous to that of Theorem 2.1.5. Consider the case of another type of performance criterion. Let

320

where

Chapter 7

are known functions from

condition

that satisfy the

and

in

Consider the pseudo-hyperbolic equation

Consider the case when By Theorem 1, the functional (12) is defined correctly and if the conditions of the theorem are valid the optimal control exists. T h e o r e m 1 3 . Consider the problem (13), (14) with righthand side Performance criterion is in the form (12). If there exists a Fréchet derivative mapping

at a point

of the

, then there exists a Fréchet

defivative of performance criterion J(h) at the same point the form


, in


P r o o f . We shall give the increment

Since the symmetry of matrix summands is equal one another.

321

to the control h*

each of the last

322

Chapter 7

By the definition of bilinear form

It is clear that the function

we have

satisfies the following equation


Since

323

then by Theorem 2.3. there

exists a solution

of the equation such that for all functions

we have

From another hand, it is clear that the right-hand side of (16) belongs to That is why by Theorem 2.3 there exists an unique solution of the problem (16), (17)

Let

and for all functions

and take into account (18)

Prove that the derivative prove this consider

is defined by expression (15). To

324

Chapter 7

Since

is a, Fréchet derivative of the mapping then

from inequality

we obtain

From another hand, applying the Lemma 1.1.3, we have

Taking into account the definition of number the inequality

it follows

we have that from


325

Let

If

min

then

Returning to (19), we have

that is what had to be proved. T h e o r e m 1 4 . If under conditions of Theorem 13 the mapping has a continuous Fréchet derivative at

the pointh* then

the

Fréchet

derivative is continuous at the point also. P r o o f is analogous to that of theorem 2.1.4. T h e o r e m 1 5 . If under conditions of Theorem 13 the mapping has a Fréchet derivative in a bounded neighbourhood of the point that satisfies Lipschitz condition with exponent then the Fréchet derivative satisfies Lipschitz condition with exponent also. P r o o f is analogous to that of Theorem 2.1.5. Consider application of the theorems in the case when the righthand side of the equation (1) is defined as As in previous sections, assume that the set is cylindrical with respect to

326

Chapter 7

space variable

and

where

is a bounded, closed, and convex set in the Hilbert space of control H. Let the right-hand side of the equation be in the form

It is easy to show that right hand side

thus

Analogously to the

it is easy to prove that the mapping

has a Fréchet derivative in the form

Prove that the derivative with exponent

satisfies the Lipschitz condition

Consider

Let v(t, x) be a smooth in numerator of the fraction

function from

Consider the



and the Schwartz inequality, we have

Apply the Schwarz inequality again:

327

328

Chapter 7

Whence,

Taking account that the set Therefore, with respect to h with exponent

is bounded, we have satisfies the Lipschitz condition Then, by the proved theorem the

performance criterion (4) has a Fréchet derivative in the domain that satisfies the Lipschitz condition with respect to h with exponent

and is in the form

. If we consider

the performance criterion J(h) directly, we prove that


In same manner we can prove that the mapping Fréchet derivative:

the derivative

329

has a

satisfies the Lipschitz condition with the exponent

and the performance criterion J(h) has a Fréchet derivative, that satisfies the Lipschitz condition with the same exponent also. Consider the other right-hand side

Prove that the mapping

Actually, we have

Whence,

has a Fréchet derivative in the form

330

Using

Chapter 7

the

inequality

we

obtain

Thus,

In the same manner, we prove that the mapping

satisfies the

Lipschitz condition with the exponent Remark that there is not any Fréchet derivative of the performance criterion (12) when the right-hand side of the equation (13) is function because the mapping has not any Fréchet derivative in the space To solve the problem, one can apply the method of regularization of control.

Chapter 8 SYSTEMS WITH HYPERBOLIC OPERATOR COEFFICIENTS

In many applied problems of science and engineering such as movement control, information transfer, radiolocation and object discovering it is necessary to solve boundary problems for differential equations with operator coefficients [141]: Such problems with certain restrictions imposed for right-hand side were investigated for the cases of some operator coefficients B in Banach spaces in the papers [78, 141]. In the cases when the right-hand side is a distribution of finite order the analogous problem was investigated in [62-64, 142, 143, 144]. The Cauchy problem when B is a generator of a semi-group was studied in [30, 78, 145]. If B is self-adjoint and positive definite operator, mixed problems for systems of differential equations containing the first and the second derivatives were been solved with the analogy of the Galerkin method in [146].

1. HYPERBOLIC SYSTEM WITH OPERATOR COEFFICIENT Let us consider a system governed by the differential equation Let B be a hyperbolic operator defined by the expression

where in some domain

are functions continuously differentiable which is determined later and supposed to be

332

Chapter 8

sufficiently smooth in

is continuous in

expression

function. The

is supposed to be uniformly elliptic in

i.e.

The functioning of the system (1) is investigated in a tube is a domain in the space of the variables bounded by the characteristic surface of the equation (1) so that the values of the component of the outward normal is positive. The reader can be convinced in the existence of such domains himself. Let us introduce the following denotations: the set of smooth in

is a completion of

functions satisfying the conditions

in the norm

is a completion in the same norm of the set of smooth functions satisfying the conditions are associated negative spaces constructed by the space of square integrable functions

and

L e m m a 1 . For any functions relation holds true

respectively. the following

SYSTEMS WITH HYPERBOLIC OPERATOR...

333

P r o o f . Let us prove the lemma for smooth functions u(t, x, y) satisfying the conditions (2). By definition of the negative norm

as far as for smooth functions u(t,x, y) the bilinear form coincides with an inner product that v(t, x, y) is a smooth in

in

We shall suppose

function satisfying the conditions

Let us consider the numerator in the right-hand side of (3)

Using the integration by parts, the Ostrogradsky-Gauss theorem and taking into account the boundary conditions, we have

Applying the integral Cauchy inequality, we obtain

Next,

Passing to the integral on the surface in the first term in the righthand side and taking into account the boundary conditions, we obtain

334

Here

Chapter 8

is the i th component of the vector of the outward normal

to the boundary

Applying to the expression

the

Cauchy inequality, we have

whence

Then, we obtain with the help of integration by parts and the Cauchy inequality that

Using obtained inequality, we find from (3) that the lemma holds true for smooth functions

Approaching the limit, we prove

that the lemma holds true for any L e m m a 2 . For any functions

the following

relation holds true

where

is the operator of the adjoint problem.

P r o o f. Let v(t,x,y) is a smooth function from

By the

definition of the negative norm, taking into account the smoothness of the function v(t,x,y), we can write


335

Let us consider the numerator in the right-hand side of the equality

Using the integration by parts, the Ostrogradsky-Gauss theorem and taking into account the boundary conditions, we have

Integrating by parts the last term in (4), we have

By the Ostrogradsky-Gauss formula, let us pass to the integral on the boundary

By the condition

Then,

For the normal vector we can write

336

Chapter 8

because the expression in the brackets coincides with the equation of the characteristics. To estimate the expression

we use the Cauchy

inequality. Then, using the integral the Cauchy inequality, we shall prove the lemma for smooth functions The final step consists in the passing to the limit. L e m m a 3 . For any function inequality holds true

the following

Proof. Let us prove that the lemma holds true for smooth functions u(t,x, y) satisfying the conditions(2). For such functions u(t, x, y) the following relation is valid where

in the domain Q. Indeed, by the definition of the function v(t,x,y),

it belongs to

Consider the functional

Using the integration by parts, the relation between u(t, x, y) and v(t, x, y) and the Ostrogradsky-Gauss formula, we can write


337

In a similar way we investigate the second term in the right-hand side of (6):

Next, for the third term in the right-hand side of (6) we obtain

Add the equalities (7), (8) and (9):

The second term in the right-hand side of (10) equals to zero, since the integrand expression coincides with the equation of the characteristics on the characteristics Hence, the relation (5) is valid. Using the Schwarz inequality, we obtain

338

Chapter 8

Reducing both parts of the inequality by

and using the

relation between u(t, x, y) and v(t, x, y) we prove that the lemma holds true for smooth functions

Passing to the limit, we prove

that the lemma holds true for any function L e m m a 4. For any

function

the

following


P r o o f . Let v(t,x,y) be a smooth function from

Let us

prove that for such functions the following relation holds true: where

in the domain Q. By the definition of the function u(t, x,y), it belongs to the space

Consider the functional

Integrating by parts and applying the Ostrogradsky-Gauss formula we obtain

In the similar way consider the second term in the right-hand side of (12):


Consider the last term in (12):

Add the expressions (14) and (15):

339

340

Chapter 8

since,

The last two expressions in the right-hand side of (16) equal to zero, since it equals to zero on y = 0 and contain the expressions, which are equal to zero on the characteristics. Adding the inequalities (13) and (16) and applying the Schwarz inequality, we obtain

Reducing both parts of the obtained inequality by

and taking

into account the relation between u(t,x,y) and v(t,x,y), we obtain that the lemma holds true for smooth functions

Passing to

the limit, we prove the lemma for any function The proven lemmas imply Theorem 1. Using these lemmas and the results of Section 1.1, we have T h e o r e m 1 . For any function there exists a unique solution of the problem (1), (2) in the sense of Definition 1.1. T h e o r e m 2 . For any functional there exists a unique solution of the problem (1), Definition 1.4.

(2) in the sense of


341

2. GALERKIN METHOD FOR DISTRIBUTED SYSTEM In this section we study the Galerkin method of approximate solving distributed system. Let the state function satisfies the boundary value problem from Section 1,

where Consider an approximate solution in the form

where

is a twice continuously differentiable function, that

satisfies boundary condition of the problem, the set total in

and the function

is

is a solution of the set of

differential equations:

where

on

the

surface

Relation (2) can be written in the form

T h e o r e m 1 . For all functions approximate

sequence

converges

the to

a

solution

342

Chapter 8

in the sense that there exists a sequence of functions

that

P r o o f . Multiplying both right and left hand sides of the equality (3) on the function summing up over i from 1 to n and integrating with respect to t from 0 to T, we have

For all functions

Assume that

let us prove the following inequality

Consider a functional

Consider the every summand of the right hand side, separately. We have

Summing up the equality over i and j from 1 to n and taking into account that

Next, we have

we obtain


Integrate the expression over

Q,

343

and take account that

Passing to the surface integrals, we find:

In the second expression we use inequality characteristic surface

Next,

Consider the following expression

Pass to the surface integration

on the

Whence, we conclude that

344

Chapter 8

We have

Consider the one part of the upper expression

By the data

and hence

On the characteristic surface the co-ordinates of the normal vector satisfy the equality

From (8) we have

Consider the other summands of (7):

It is easy to see that on the characteristic surface the integrated function equals to zero.


345

By this reasons, the inequality (6) is valid for all Passing to the limit, we prove the inequality for all functions By the inequality (6) and the Schwarz inequality

dividing by

we have

From the inequality we conclude that there exists a weakly convergent subsequence

Let the subsequence

converges weakly to a function

By the Banach theorem

there exists a subsequence converges to the same function

that the sequence in norm of the space

Multiplying the both right- and left-hand sides of (3) on a function and integrating with respect to t from 0 to T, we have where Consider fundamental sequence

By Lemma 1.6.1. we have

346

Chapter 8

and hence

Since the sequence


there exists a limiting function

Show that function

in the space

of the sequence

By (9) and the definition of

we have the following equality


The right-hand side of the inequality vanishes as we have

From (10)

Since the number i in (11) is arbitrary and the set of functions is dense in the space we obtain That is why

and therefore the function u(t, x, y) is a solution of the problem in the sense of Definition 1.1.1. Since the imbedding operator of the space in subsequence

is compact, the chosen weakly convergent converges in the space

hence,

u(t,x, y) is a solution of the problem in the sense of Theorem 1. Remark. There is no necessity to choose a subsequence because it follows from the results of Chapter 1 that there exists a unique solution of the problem.


347

3. PULSE OPTIMAL CONTROL OF THE SYSTEMS WITH HYPERBOLIC OPERATOR COEFFICIENT Consider the problems of impulse optimal control (Section 3.6) for the systems with hyperbolic operator coefficient

The denotation is the same as in Section 6.1. Consider the initial boundary value problem

Complete the table of the template theorems of Section 3.6.

348

Chapter 8

Chapter 9 SOBOLEV SYSTEMS

In the chapter we study systems that don’t satisfy the conditions of the Cauchy-Kovalevskaya theorem. These systems were investigated in [147-150].

1. EQUATION OF DYNAMICS OF VISCOUS STRATIFIED FLUID Many applied problems (such as oceanographic research, oil barging, etc.) are described by the dynamic equation of viscous stratified fluid. In the papers [151, 152] the equation of dynamics of plane motion of viscous exponential stratified fluid was obtain by Boussinesk approximation:

where is the Laplacian. In [152] the solvability of an initial boundary value problem for this equation is studied. There were shown the existence and uniqueness of a generalized solution with the right-hand side from the negative Hilbert space with respect to variable. In this section we study the optimal control problems of the generalized dynamic equation of viscous stratified fluid with the righthand side from the negative Hilbert space with respect to and variables. Consider the system

350

in the tube boundary

Chapter 9

where

is a regular domain with

is the co-normal derivative,

the normal vector to the surface

is is a matrix of

elements Operators expressions:

and

are defined by the formal

with sufficiently smooth coefficients in a closed domain . We assume that the coefficients satisfy the following conditions: . for-all Assume also that for all real

where

are positive constants.

SOBOLEV SYSTEMS

Let

351

be a space of measurable, squared integrable functions,


functions, that satisfy

conditions (2) in the norm


functions, that satisfy

adjoint conditions

in the same norm (3). Let


Prove the following lemmas. L e m m a 1. For any function

the following


P r o o f. For smooth functions (2), we study the following expression

where

that satisfy the conditions

is a bilinear form that is defined on

Applying integration by parts and passing to the surface integrals, we have

352

Chapter 9

Estimate each summand in the right-hand side of the inequality (6). Taking into account the Schwarz inequality, we obtain

To estimate the second summand of (6), we use the Schwarz inequality and coefficient boundedness of the operator B :

In the same manner, we show that

Taking into account proven inequalities and (5), we obtain the desired inequality for smooth functions Passing to the limit, we have the inequality for all functions In the same manner we prove the following assertion

SOBOLEV SYSTEMS

L e m m a 2 . For any

353

function

the

following

the

following


where

is the adjoint operator

L e m m a 3 . For any inequality holds true

function

P r o o f. Show the desired inequality for smooth functions that satisfy the conditions (2). Consider the auxiliary operator

where

For the functions

prove the inequality

Since the function satisfies the conditions (2), then the function satisfies the conditions (4). In addition the functions satisfy the following relation Estimate the expression

354

Chapter 9

Employing the integration by parts and passing to the surface integrals, we obtain

where

Thus,

SOBOLEV SYSTEMS

355

Using inequality of coercivity [74] for uniformly elliptic operator B, it is not difficult to prove, that

From the estimate of the norm of the elliptic operator B , we have

Applying to the left-hand side the Schwarz inequality, we find

then

From the relation between

and

Thus, for all smooth functions the inequality in Lemma 3 is proved.

we obtain

that satisfy the conditions (2),

To prove the inequality for all functions

we employ

passing to the limit.

L e m m a 4 . For any function

the following

inequality holds true P r o o f of Lemma 4 is much analogous the previous proof. The auxiliary integral operator is of the form

where

356

Chapter 9

T h e o r e m 1 . For any function there exists a unique solution of the equation (1), (2) in the sense of Definition 1.1.1. T h e o r e m 2 . For any function there exists a unique solution of the equation (1), (2) in the sense of Definition 1.1,4.

2. ONE SOBOLEV PROBLEM Consider a function differential equation:

that is a solution of the following

with boundary conditions

where A, B are differential operators:

where

are continuously

differentiable functions in the domain The differential operators (3) satisfy the following conditions:

We study the system (1), (2) in a tube

.

SOBOLEV SYSTEMS

357

Let be a completion of the set of smooth in which satisfy the following conditions

functions,

in the norm


functions, which

satisfy the adjoint conditions in the same norm (5),


L e m m a 1 . For any function inequality holds true

the following

P r o o f. First prove the inequality for smooth functions which satisfy the conditions (2). By the negative norm definition,

Applying integration by parts and passing to the surface integrals, we have

358

Chapter 9

Employing the Schwarz inequality, we obtain

In the same mañner estimate the other summands of L :

Substituting these inequalities into (7) and dividing by

, we

obtain the desired inequality for smooth functions which satisfy the conditions (2). Continuously extending the operator L to the space we have the inequality for all functions. L e m m a 2 . For any function

the following

inequality holds true Lemma 2 is proved in much the same manner as Lemma 1. From the proved inequalities it follows that the operators L and can be extended to the continuous operators mapping and into

and

, respectively.

L e m m a 3. For any function inequality holds true

the following

SOBOLEV SYSTEMS

where

359

is a completion of the set of smooth in

functions,

which satisfy the conditions (2), in the norm

P r o o f . First prove the lemma for smooth functions satisfy conditions (2). Consider the auxiliary functions of the form:

It is obvious that

that

. Prove the following inequality:

Employing the integration by parts and taking into account relation between and , we obtain

Passing in the formula (9) to the surface integration and taking into account boundary conditions, we have

360

Chapter 9

In the same way consider the other summands

Adding (10) and (11), we obtain (8). Applying the Schwarz inequality to left-hand side of (8), we have desired inequality for smooth functions Passing to the limit, we obtain the inequality for all functions L e m m a 4 . For any function

the following


where

is a completion of the set of smooth in

functions,

which satisfy the conditions (6), in the norm

The operators and L are defined by the same differential expression but Lemma 4 is proved in much the same way as previous one. The auxiliary function is in the form

T h e o r e m 1 . For any function

there exists a

unique solution of the equation (1), (2) in the sense of Definition 1.1.1. T h e o r e m 2 . For any function there exists a unique solution of the equation (1), (2) in the sense of Definition 1.1.4.

SOBOLEV SYSTEMS

3. PULSE OPTIMAL CONTROL OF THE DYNAMIC EQUATION OF VISCOUS STRATIFIED FLUID Complete the templates of the theorems of Section 3.6

361

362

Chapter 9

SOBOLEV SYSTEMS

4. PULSE OPTIMAL CONTROL OF ONE SOBOLEV SYSTEM Using the templates of Section 3.6, complete the tables.

363

364

Chapter 9

Chapter 10 CONTROLLABILITY OF LINEAR SYSTEMS WITH GENERALIZED CONTROL

1. TRAJECTORY CONTROLLABILITY In investigations of various dynamic systems the problem of its controllability is one the most important. The problem of the controllability was studied in [6] for the linear ordinary systems, which allow the generalized controls. It was pointed out there that the introduction of such controls does not extend the conditions of the complete controllability. In the case of distributed systems the situation is much more difficult. It was shown in the paper [14] that the solvability of the problem of the controllability for distributed systems with a point control significantly depends on the fact whether the point of the control application may be approximated by the Diophant approximations well enough. Various conditions of the controllability of lumped and distributed systems were obtained, for examples, in [30-32, 84, 86, 153-157, 177]. Let be a solution of the problem (1.1.1.12). By controllability of the system (1.1.1.12) we shall mean the possibility to reach any state u(t, x) as a result of admissible controls To investigate the controllability, it is necessary to study the properties of the operator L . This problem is solved with the help of the apparatus of equipped Hilbert spaces and inequalities in negative norms [63, 122, 142, 143, 158-162, 177], and also with the help of correspondent inequalities in positive norms, which follow from them.

D e f i n i t i o n 1 . The system is controllable in a Banach space W by the set of admissible controls if the set covers

the

space in W.

W,

i.e.

366

Chapter 10

D e f i n i t i o n 2. The system is space W by the set of admissible controls is dense in such that

in the Banach if the set W, i.e.

Let the following inequalities in the negative norms still hold true for

The inequalities in the negative norms and the results of Section 1 imply the following theorems. T h e o r e m 1 . If operator maps surjectively into the whole space

then the system (I.1.I2) is controllable

in the space P r o o f . Let

be an arbitrary element belonging to

Due to the inequalities (2), we obtain that the operator

surjectively maps

such element

As far as into

there exists

that A(h *) = Lu * (t, x) – f in

Hence, by Lemma 1.3, we obtain that

where hand side

is a solution of the problem (1.1.12) with the rightin the sense of Definition 1.1.4. For this reason, in

and hence, in

also. Taking into

CONTROLLABILITY OF LINEAR SYSTEMS ...

account that

367

we conclude that

It

follows from Lemma 1.4 that is a solution of problem (1.1.12) in the sense of Definition 1.1 also. Thus, the system is controllable in T h e o r e m 2. If operator that the set

maps

is dense in

in

so

then system (1.1) is

in P r o o f . Let u* (t,x) is an arbitrary element belonging to Since

is densely imbedded in such

there exists a sequence

that

Consider

with fixed k. As far as there

exists

a

sequence

of

is dense in

controls

such

Choose such

that that

be a solution of the problem (1.1.12) with the right-hand side sense of Definition 1.1.4. Granting Lemma 1.3 we obtain

In other words, the system is

in

in the

368

Chapter 10

It is well known [78], that the linear span of the functions is dense in the set

and hence in

also. On the other hand, it

is easy to prove that the sequence

converges to the function decomposition Thus, the set

is dense in

as the diameter of the of the interval (0,T) tends to zero.

and hence, the previous theorems are

valid in the case of pulse controllability.

2. TRAJECTORY-FINAL CONTROLLABILITY OF SOME LINEAR SYSTEMS The problems of controllability of linear systems have one essential difference from the classic problems for systems with concentrated or distributed parameters. In this problems it is necessary to provide the required state of the system with the help of controls during the whole time interval of the system functioning (trajectory controllability) rather than to transfer the system into a desired state in a finite time interval (final controllability). It is naturally to consider the problem of the final controllability in a finite time interval in the case of linear distributed systems with generalized controls. The singularity of the right-hand side makes this consideration difficult enough, since the function u(T,x), may


369

does not belong to but it may be an element of some negative space. However, in the case of hyperbolic and pseudo-hyperbolic systems it is possible to obtain a positive results concerning the final controllability in a finite time interval. Moreover, it is possible to provide the desired state of the system with an arbitrary given precision (in an integral metrics). Thus, we may say about the trajectory-final controllability. Let us consider in detail the system described by the Newton pseudo-hyperbolic initial boundary value problem. Let be a bounded domain in with a regular bound The state u(t,x) of the considered system is a solution of the following problem

where the parameter h is a control from the set of admissible controls , V is a space of controls. Remark. The following results will be valid also if we shall replace the Laplacian by symmetric elliptic operators of the second order with smooth coefficients and the conditions (3) will be replaced by the corresponding Dirichlet or Neumann conditions. Next, we shall investigate the generalized solvability of the problem (l)-(3), basing on which we can prove the trajectory-final of the system.

370

Chapter 10

2.1. Generalized solutions and Denote by a completion of D(L) (the set of smooth in functions satisfying the conditions (2), (3)) in the norm

is a completion of and

in the same norm. Here is the set of smooth in

functions satisfying the adjoint conditions

Let us construct on the space

the negative spaces

using the positive spaces introduced above Introduce the pair of the spaces H,

as completions of D(L),

in the following norms

The spaces H, spaces

are isometric to the direct sum of the Hilbert and there are no imbeddings


371

Lemma 1. For any function following inequalities holds true

the

Proof. Let us prove the inequality (4) ((5) can be proved in the similar way). The right-hand side, of the inequality (4) is proved by applying of the integration by part, the Ostrogradsky-Gauss formula and the integral Cauchy inequality to Here v is an arbitrary smooth in function satisfying the adjoint boundary conditions. To prove the left-hand side of (5), consider the expression

where

is an arbitrary smooth in

function satisfying the

boundary conditions (3), (4). Note, that Indeed, the initial condition

holds true as a result of the

construction of the auxiliary operator. The others boundary conditions disappear after completion of in the norm Expressing u in terms of Iu , we obtain We have where

Consider every term

separately.

372

Chapter 10

Using the integration by parts and taking into account the conditions

and the fact, that

we have

Using the Ostrogradsky-Gauss formula and taking into account the boundary condition

Consider every term separately. a)

we have


373

b) To estimate the second integral, we must use the integration by parts and the conditions We obtain

Granting that

we can write

c) Let us show, that the third integral in nonnegative. To do this, we must apply the integration by parts.

Collecting together all the obtained inequalities and taking into account, that we have

374

Chapter 10

Applying to the right-hand side the inequality reducing

by

and we

have

Granting that

we have that (4) is valid. The lemma is proved. Basing on the right-hand sides of the inequalities (4), (5), we can extend with respect to continuity the operator L ( , respectively) onto the whole space

respectively) and consider that it

continuously maps into

respectively). We save the same

denotations for the extended operators. The inequalities (4), (5) are valid also for the extended operators for any Using inequalities obtained above in the similar way, it is possible to prove the unique dense solvability of the operator equations

Let

be the negative spaces constructed on

the corresponding positive spaces

and

which, in turn, are


completions

of

D(L),

in The

375

the

norm

imbeddings

are valid and dense. The following theorems hold true.

Theorem 1. For any element solution


of the equatioh Lu = f .

Theorem 2. For any element solution


of the equation

Let us define a generalized solution of the problem (1), (2), (3) as a function for which there exists a sequence of smooth functions m = l,2,..., such that

It should been noted, that the similar conception of a generalized solution of the operator equation with closed linear operator in a Banach space was proposed in the paper [165]. Theorem 3. For any element generalized solution Proof. Let


of the problem (1)-(3). By virtue of the density

there exists such sequence

that

By Theorem 1 for any function solution

in


of the operator equation

hand side of the inequality (4) and the imbedding

Using the leftwe have

376

Chapter 10

Thus, by virtue of the completeness of the space H there exists such function that Let us show that density D(L) in

is desired generalized solution. From the it follows that for any natural number m there

exists a sequence of smooth functions that

Choose

i = 1,2,..., such from the condition

On the other hand,

The uniqueness is proved in the standard way by contradiction. The theorem is proved. D e f i n i t i o n 1 . The system (l)-(3) is in the space H by the set of admissible controls if the set is dense in H. T h e o r e m 4 . If the set

is dense in

then the

system (1)-(3) is in the space H. P r o o f . Let an arbitrary function be given. Consider the space W, which is a completion of the set D(L) in the norm It is possible to prove that the imbeddings are valid and dense, and in addition, the following inequalities hold true


377

Moreover, it follows from the proof of Theorem 3 that generalized solutions belong to and As far as W is dense in H, there exists such sequence that

By virtue of the density of

in

for any natural number i there exists such sequence of the admissible controls

that

Then, we have

The theorem is proved. In this case the imbedding

is valid

and dense, so we can conclude that the system (l)-(3) is trajectoryfinal controllable in the class of pulse controls.

3. PULSE CONTROLLABILITY OF PARABOLIC SYSTEMS Next, we shall consider some other problems of controllability on the examples of parabolic systems with impulse impact. Urgency of such problems is stipulated by both the arising of new technologies and the simplicity of the control on the basis of spatially distributed impulse of given class with simply regulative control. For the system with continuous and discrete control similar problems were posed and investigated in [161, 162].

378

Chapter 10

Let

the

functioning of the system in a tube with a regular boundary is described by the

equation

where B

is uniformly elliptic operator in The control is made with the help

of coefficients

are moments of impulse

impacts

In consequence of the

presence of Dirac with respect to the time variable in the right-hand side of the equation (1) the solution of the problem (1), (2) can be represented in the following form:

where

is

the solution of the problem (1), (2) with the right-hand side

Obviously,

when

and satisfies the identity

It follows from the relation (4), that in is a solution of the problem

the function


which, as well-known, belongs to the space function the point

379

Hence,

is continuous on the segment [0,T] everywhere, except and moreover, it is continuous from the

right. The solution of problem (5) we may represent in the form [166]

where is a semi-group generated by the operator B . Under the above mentioned restrictions the operator B generate an orthonormal basis in which consists of the eigenfunctions

and,

moreover, only finite number of the may correspond to each eigenvalue

For any

The solution the of problem (1), (2) at the moment we may represent in the form

eigenfunctions

380

Chapter 10

D e f i n i t i o n 1. The system is impulse controllable if the set is dense in where D e f i n i t i o n 2 . The system is impulse controllable at N steps in if the set where is dense in G T h e o r e m 1 . In order that the system (9), (10) be controllable in the sense of Definition 1, it is necessary and sufficient that for any n = 1,2,..., the following conditions is satisfied P r o o f . Sufficiency. On the basis of the criterion of the density in the system (1), (2) is impulse controllable if for it follows from the equality that z = 0. Granting the representation (8), in particular, we have that

This implies that

Under (hence, side in (12) tends to zero. Hence,

the second term in the left-hand

In the similar way, we can show that for any n = 1,2,... we have


381

If condition (9) is satisfied then from (13) we obtain that Hence, z = 0 . Necessarily. Let the system (1), (2) is controllable in the sense of Definition 1 but for some n the condition (9) is not valid. Then there exists a nonzero element such that the relation (11), and hence (10), hold true, that contradicts to Definition 2. T h e o r e m 2. The system (1), (2) is not controllable in in the sense of Definition 2. This result follows from the fact that under finite number of impulses N the system of linear algebraic equations (11) with

respect

to

the

variables

n = 1,2,..., has infinite number of nonzero solutions. Thus, the system is not controllable. Denote by a subspace generated by the eigenfunctions of the operator B corresponding to the first eigenvalues, and let

be an

approximate solution of the problem (1), (2) obtained by the Galerkin method, where the same functions were taken as a basic ones. The, as well-known [142] , strongly in The functions

at the points

T h e o r e m 3 . In controllable in

order

have the form

that the system (1), (2) be in the sense of Definition 2

it is necessary and sufficient that the condition (9) holds true for any

382

Chapter 10

P r o o f . Sufficiency. It is obvious, that

is an arbitrary element from

The system (1), (2) will be controllable in the sense of Definition 2 if it follows from the equalities that

Granting (14), it follows from (15) that, in particular,

Relations (16) form the system of linear equations where

and matrix

has block form

Introducing the notation

we have

Solving the system (17) by the Gauss method, we transfer the matrix to the following form:


where

are

383

some

constants

and

do not equal to zero. Indeed, to prove this fact we may just show that the determinant

does not equal to zero. It is easy to see that the determinant F is a Vandermond determinant with respect to which does not equal to zero as far as

when

Granting (9) and (18) we can conclude that the rank of the matrix and hence

is equal to

and coincide with the dimension

of the vector It implies that the system (17) can have only zero solution. Necessarity. Suppose that for some n the condition (9) is not valid. Hence, the rank of the matrix of the system of linear algebraic equations (17) is less than the number of variables. It means that there exist the nonzero solutions of the systems. Thus, there exists such that (15) is valid, and hence the system (1), (2) is not controllable.

384

Chapter 10

4. SUBDOMAIN CONTROLLABILITY OF PSEUDO-PARABOLIC SYSTEMS Consider a pseudo-parabolic system with controls concentrated in subdomains of the space domain. R e m a r k . The result represented below holds true also for the case of point control in a one-dimensional space domain. . For hyperbolic equations one of the ways of the investigation of the point controllability on the basis of the concept of “exact controllability” described in the J.-L.Lions' work [31, 32]. Let the state of the system is a solution of the initial boundary value problem

where

is the indicator of the subset (subdomain)

positive Lebesque measure,

with

is a given "elementary"

intensivities. The control of the system is carried out with the help of the vector-function The differential expression A has the form

It is uniformly elliptic in is a positive constant. By virtue of the restrictions introduced above the differential expression A yields in the space a symmetric and positive definite operator with a discrete spectrum, namely: there exists a


385

countable set of the eigenvalues which can be arranged in ascending order: and a complete orthonormal in eigenfunctions

system of the corresponding and only finite number of linearly

independent eigenfunctions

can be assigned to

every eigenvalue The generalized solution the problem (1), (2) belongs to the space which is the space of continuous on [0, T] functions taking on values in the Sobolev space

It follows from the

imbedding of the space into the space With the help of the Fourier method the solution of the problem (1), (2) can be represented in the following form

It is requested to transfer the system to the desired state before the time moment T with the help of the controls h(t). The controllability of the system (1), (2) we shall mean in the sense of the following definitions. D e f i n i t i o n 1 . The system is a subdomain controllable one if the set is dense in T h e o r e m 1 . The system (1), (2) is a subdomain controllable one in the space if and only if for an arbitrary natural number n the following condition is satisfied where

386

Chapter 10

P r o o f . At first, let us prove the sufficiency. Let us show that for any the condition implies that we have

Granting the representation of the solution (3),

Whence,

It follows from (6) that

The

function

analytic in the right-hand complex half-plane to zero in the interval (0, T ). Thus, on as in the proof of Theorem 3.1, we obtain that

is and it is equal Reasoning


387

Denote

We can rewrite (8) in the form valid,

n = 1,2,.... Since (4) is

n = 1,2,.... Thus,' z(x) = 0.

Let the system (1), (2) is controllable in the sense of Definition 1 but for some natural number k the condition (4) does not satisfied. The there exists such vector that Let the function

be such that

We have that for any n

Whence, we obtain that contradicts to Definition 1. The theorem is proved. The following example is sufficiently instructive. The system

is controllable if is rational number

is irrational number and is not controllable if

388

Chapter 10

R e m a r k . For the case when Theorem 1 can be generalized for the systems with point controls

in the following way: the system '(9), (10) is controllable if and only if for an arbitrary natural n the following condition holds true where

Here

is

an

eigenfunction

of the

operator

A

corresponding to the eigenvalue The condition is equivalent to the condition that not all components of the vector equal to zero.

Chapter 11 PERSPECTIVES

1. GENERALIZED SOLVABILITY OF LINEAR SYSTEMS 1.1. Basic definitions and facts Concept of a generalized solution of an operator equation with an arbitrary linear closed operator L was proposed in [165]. Recall the basic principles of this approach. Let E, F be Banach spaces and L be a closed linear injective operator mapping E into F with everywhere dense in E domain of definition D(L) and with dense in F range of values R(L). In a parallel way (1), we shall investigate the adjoint equation where

is the adjoint operator.

Suppose also that the set

is total in F ,and the set

is total in E in duality The condition of totality of may be replaced by one of the following assumptions: A) the kernel of L consists of zero-vector

and

is weakly dense in E (this condition independently implies the injectivity of L and the density of R( L) in F ); B) the operator L is continuous (D( L) = E) ; C) the space E is reflexive (if in addition F is reflexive then the set

is total in F also).

Under the assumptions A) and C) the totality follows from the well-known results, and in the case B) it follows from the formula

390

Chapter 11

where is the polar of the set in duality (E, E*). Let us prove that (3) holds true for an arbitrary closed linear operator

as far as D(L*) is a total linear subspace then Thus, the formula (3) is proved. It follows from (3) that

If L is a continuous injective operator then Ker(L)={0}. Therefore,

D(L) = E,

Thus, the bipolar of the set R(L ), i.e. the weak closure of R(L ), coincides with E , that implies the totality of R(L ) in E Denote by

the completion of the set E

in topology

as far as the sets E and R(L ) are in duality then is a separable locally convex linear topological space. Consider an arbitrary linear functional

The equation (1) implies that

By the Banach theorem about weakly continuous linear functional [78], the functional allows the unique extension by continuity on the whole space D e f i n i t i o n 1. The generalized solution of the operator equation (1) is an element obeying the relation

PERSPECTIVES

391

D e f i n i t i o n 2 . The generalized solution of the operator equation (1) is such element that there exists a sequence (which we shall call as almost solution) obeying the relations The concept of a generalized solution arises when the right-hand side of the operator (1), i.e. the element does not belong to the range of values R ( L ) of the operator L (in this case there are no any classic solutions). But if than the generalized solution becomes a classic one. In [165] the following theorem was proved. T h e o r e m 1 . For any element there exists a unique element which is a generalized solution of the equation (1) in the sense of Definitions 1,2. Let us introduce once more definition of a generalized solution. Let is a completion of a linear set D(L) with the norm The equality

allows extending L by

continuity from the set D(L) of the space E onto the whole space The extended operator we shall denote by The operator is linear and continuous. D e f i n i t i o n 3. The solution of the equation (1) is such element that the equality (1) holds true for the extended operator It is easy to see that the extended operator determine an isometric isomorphism between the spaces and F . T h e o r e m 2. For any element there exists a unique generalized solution of the equation (1) in the sense of Definition 3.

392

Chapter 11

As well-known, the case of closed linear operator L reduced easy to the case of a linear continuous operator L . Indeed, introducing the norm of the graphics on D( L) relatively to which the linear set D( L) is a Banach space, we obtain that the operator is linear and continuous Taking into consideration the mentioned above, we shall further investigate only the case of continuous linear operator L (D(L) = E). At first, we shall determine the relations between the solvability in the sense of Definitions 1,2 and 3. T h e o r e m 3 . The

space

is

algebraically

and

topologically imbedded into the space

P r o o f . Let some net topology of the space hence,

converges to 0 in the Then

for any

Thus,

Whence, we have that the topology It still remains to prove that if space

in the space F, and

in the topology

is weaker than the topology in the topology of the then u = 0 (the condition

It follows from the condition

in

and whence

that

implies that we

have

that

By the totality of the set D(L ) and the injectivity of the operator we have that u = 0. Thus, T h e o r e m 4 . Definitions 1,2 and 3 are equivalent.

PERSPECTIVES

393

P r o o f . Let us prove that the solution in the sense of Definition 3 is a solution in the sense of Definition 1 also (inverse statement is obvious). Indeed, in

there exists a solution

of

the same equation Lu = f . It is clear that Whence we have that u = Ou is an operator of imbedding of the set into Often constructive description of the spaces and is sophisticated and in general case still be an unsolved problem. Therefore it is necessary to find out an algebraic and topological dense imbedding of the space or into some other well-known Banach or locally convex linear topological space H . In this section we shall describe such spaces H . Besides, we shall investigate the properties of the generalized solutions in these spaces H .

1.2 A priori inequalities Taking into account the mentioned above, suppose that the space is algebraically and topologically imbedded into the Banach space H . By the condition of the topological imbedding Whence, we conclude that where is positive constants. The estimations of such kind are usual in applications and they are called a priori inequalities. Except for (6), it is easy to see that the following a priori inequalities hold true

394

Chapter 11

Note that the inequalities (6) themselves do not guarantee the existence of a topological imbedding , but it only compare the topologies induced by the norms

in the set E .

Hereinafter, we shall prove that the estimations (6) can serve as a basis of the theory of the generalized solvability of the operator equation (1).

1.3. Generalized solution of operator equation in Banach space It follows from the results of the previous subsection that the inequalities (6) are necessary conditions for the construction of the theory of generalized solvability of the operator equation in the Banach space H . We shall prove that the inequalities (6) are also sufficient conditions of the solvability of the equation (1) in some generalized sense. Note that proposed scheme cover various approaches to the construction of generalized solutions of differential equations (see [167], for example). Thus, we shall suppose that the linear operator satisfy the inequalities (6) where H is a completion of the space E with the norm . It is clear that the. right-hand side of the inequalities (6) implies the continuity of the operator L, and the left-hand side implies the injectivity. Besides, by virtue of the density of the imbedding the set H* is total in E* and, hence, the spaces E and H are in duality. Consider the operator defined in the following way:

L e m m a 1. The operator equation subset H of the space E .

is solvable in the

PERSPECTIVES

395

P r o o f . The left-hand side of the inequalities (6) implies the correct solvability of the operator Consider also the adjoint operator

It is clear that if then

functional

where

is a contraction of the

from the set H onto E . As well known, the

correct solvability of the operator implies the everywhere solvability of the operator [165], whence, taking into account the mentioned above, we obtain the solvability of the operator L on the set H (considered as a subspace of E ) of the right-hand sides in (2). R e m a r k 1 . If the operator L satisfies the inequalities (6) then D e f i n i t i o n 4. A generalized solution of the equation (1) with the right-hand side is such element that the equality holds true for any It is obviously that the equality (7) is equivalent to T h e o r e m 5. For any right-hand side there exists a unique solution of the equation (1) in the sense of Definition 4. P r o o f. Choose a sequence such that in the space F . If

is a solution of the equation

granting (6) and the fact that the sequence have

then

is fundamental, we

Chapter 11

396

Thus, there exists such

in the space H .

Further, we have Passing in the last equality to the limit as

we obtain

Thus, is a solution of (1) in the sense of Definition 4. As long as the equality implies that u* = 0, and hence, the solution is unique. D e f i n i t i o n 5 . A generalized solution of the equation (1) with the right-hand side is such element that there exists a sequence satisfying the conditions T h e o r e m 6. Definitions 4 and 5 are equivalent. P r o o f. Let be a solution of the equation Lu = f in the sense of Definition 4, i.e.

Reasoning similarly to

Theorem 5, we conclude that u = u , and hence, the other hand,

On Thus,

is a solution

of the equation (1) in the sense of Definition 5. Let us prove the inverse assertion. Let be a solution of the equation (1) in the sense of Definition 5. Then,

for any Estimate the first and third summands in the right-hand side

PERSPECTIVES

397

Whence we have that i.e. u is a solution of the equation (1) in the sense of Definition 4. R e m a r k 2. It is obviously that the solutions in the sense 4 and 5 coincide with a classic solution It is also easy to testify that in this case the classic solution is a generalized one. In the case when the generalized solution u belongs to D( L) it is a classic one, T h e o r e m 7. If the space is continuously imbedded into H then Definitions 3, 4, and 5 are equivalent. P r o o f. Let us prove that Definition 3 is equivalent to definition 5. Let be a solution of (1) by Definition 5. As it was indicated above, for any right-hand side there exists a unique solution in the sense of Definition 3. Considering u as an element of the space H (by virtue of the imbedding

we can prove

that u = u. Indeed,

where is a sequence defining the solution Thus, u is a solution of (1) in the sense of Definition 3. Let now, conversely, u be a solution of (1) by Definition 3. Choose an arbitrary sequence such that Then by the continuity of the imbedding

we have

398

Chapter 11

That is what had to be proved. R e m a r k 3. The theorem implies that there exists a constant c> 0 such that the following inequality holds true: where u is a solution of (1) with the right-hand side f in the sense of Definition 3, 4, and 5. R e m a r k 4. The imbedding follows from the density of

in the space F in the weak topology

or from the fact that the operator is closable. Usually from the inequalities (6) for the operator L it is possible to prove the similar inequalities for the adjoint operator where G is a completion of the set F with some norm. Consider this situation for reflexive Banach spaces E, F . In this case and similarly to Lemma 1 we obtain that the operator equation (1) is solvable in . Besides, we have the theory of solvability of the adjoint equation (2) (analogous of the Theorems 5, 6, and 7).

Theorem 8. There exists a constant c> 0 such that for any and any the following inequalities hold true

where

are solutions of the equations Lu= f and

L u = l. P r o o f. Let us prove the inequality (9) (the inequality (10) is proved in the similar way). As far as the equation (2) is solvable (in the sense of the analogous of Definitions 4 and 5 for the adjoint operator)

PERSPECTIVES

399

for any then for all the following equality holds true (the second adjoint space is identified with the original space) where is a solution of (2) with the right-hand side Whence, we have

or

Thus, the set of functional

is bounded in every point , and hence, by the Banach-Steinhaus theorem it is bounded by the norm of the space E = E , hence, the inequality (9) is proved. T h e o r e m 9. There exists a constant c> 0 such that for any and any the following inequalities hold true

where

are solutions of the equations Lu = f and

L u = 1 in the sense of Definitions 4 and 5. P r o o f. Granting the inequality (11), we have Applying to the right-hand side the inequality (9), we obtain

or

400

Chapter 11

Taking into account that we obtain the inequality (13). The inequality (12) is proved in the similar way.

R e m a r k 5 . As far as

and the

inequality (12) holds true for any

it may be seemed that

(12) provides the existence of imbedding is not true. In the space

really exists an element u such that

But if the imbedding cannot to compare Definitions 4 cnd 5 and of Definition 3.

However, this

is not proved, we

as a solution of (1) in the sense of as a solution of (1) in the sense

1.4. Generalized solutions in locally convex linear topological spaces Let us introduse several more definitions of a generalized solution. As before, we suppose that L is a linear continuous operator. Select in the space E some total linear set

Let

is a

completion of the set E by topology By the Banach theorem about weakly continuous linear functional [78] the functional allows a unique extension by continuity on the whole space D e f i n i t i o n 6 . A generalized solution of the equation (1) is an element obeying the relation

PERSPECTIVES

401

D e f i n i t i o n 7. A generalized solution of the equation (1) is an element for which there exists a sequence such that in topology

as

In the case when M = R(L ) Definitions 6 and 7 coincide with Definitions 1 and 2 of a weak solution and an almost solution. As in [165], we can prove the following

T h e o r e m 1 0 . Definitions 6 and 7 are equivalent and for any element there exists a unique generalized solution of the equation (1)in the sense of Definitions 6 and 7. R e m a r k 6.

Definition 7 implies that

if

is a

generalized solution of (1) then the point (u,f) is an point of accumulation of the graphics set

of the operator L, i.e. of the

in the topology Remark 7. Moreover, the point

(u, f)

belong to the

sequential closure of the graphics in the topology , and hence, is a sequential closure of the set E in the topology Taking into consideration Remark 7, we could investigate the concept of generalized solution in sequentially complete spaces In addition, it is easy to prove that a classic solution of (1) is a generalized one in the sense of Definitions 6 and 7. If or the generalized solution belongs to D( L) then the generalized solution becomes classic. As long as

then there exists such set

that

402

Chapter 11

T h e o r e m 11 . In order that there exists an algebraic and topological imbedding of the space into the space it is necessary and sufficient that be total linear subset of the space F . In this case Definitions 6, 7 and 3 are equivalent. P r o o f. Similarly to Theorem 3 we can prove that on the set E the topology induced by the norm stronger than the topology In the similar way we consider the condition that

We have

By virtue of the fact that the operator

determines an isometric isomorphism between the spaces and F , we conclude that the condition u = 0 is equivalent to the totality of the set The equivalence if the definitions is proved similarly to Theorem 4.

R e m a r k 8. Theorem 11 implies that the space can be introduced (under the assumptions of the theorem) as a completion of the set by the topology R e m a r k 9. Note that the condition of the totality of has the principal character and not always holds true. It is easy to give an example of a linear continuous injective operator mapping not total sets into total. Indeed, the operator

acting on the vector

according to the rule

is obviously injective, linear and continuous continuous) operator that maps the vector system

(even completely

PERSPECTIVES

403

into the system

The vector system

is orthogonal to the vector

and, hence, is not total, on the other hand the totality of the system in the space is obvious. Besides, if we suppose that the operator L obeys the a priori inequalities (6), this implies the following T h e o r e m 1 2 . The Banach space H is algebraically and topologically imbedded into the space . Definitions 6 and 7 for the set and Definitions 4 and 5 , respectively, are equivalent. P r o o f. It is clear that the topology induced by the norm stronger than the topology the net

Let us test the condition

Let

converges to u in the topology of the space H and in the topology

theorem the norm

Then by the Hahn-Banach

may be represented in the form

therefore,

On . Thus,

the

other

hand,

and hence, u = 0 .

404

Chapter 11

R e m a r k 10 . The concept of a generalized solution investigated in [167] is equivalent to the concept considered in this subsection.

1.5 Connection between generalized solutions in Banach spaces and in locally convex spaces Note that the constructions of generalized solutions in the sense of Definitions 4 and 5 and constructions of generalized solutions in the sense of Definition 6 and 7 are analogous in some sense. On the basis of the total set

it is possible to

construct the Banach space giving a generalized solution in the sense of Definitions 4 and 5 in a way analogous to that used in the construction of a separable locally convex linear topological space on the basis of the total set. Namely, let is a completion of the set E by the norm

The norm (14) can be rewritten in the form

The spaces induced by the norms (15) were investigated in detail in the case of total sets On the other hand if M = R(L ) then the norm with the norm of the space analogy of the space

, and hence, in this case Indeed,

coincides (an

405

PERSPECTIVES

The last equality holds true, since by the Hahn-Banach theorem for any element there exists such functional with the unit norm L e m m a 2. If space F then

and

is total subset of the

algebraically and topologically imbedded into

the space P r o o f. Totality of the set

and injectivity of the operator L

implies the totality of the set M . In is easy to see that the norm stronger than the norm Let the sequence also

It still remains to testify the condition converges to

in the norm

converges to zero in the norm

then on the one hand

but on the other hand Whence,

for any

By virtue of the totality of

and injectivity of we have u = 0. R e m a r k 11. The theorem implies that assumption of the totality of the space constructed by completing the set

and

under the can be

with norm (14).

Thus, on the basis of any total set it is possible to obtain a priori estimation (under the assumption of the totality of we can prove this estimation even with the imbedding which implies the assertions about solvability of the equation (1) (in the sense of Definitions 3, 4, and 5) similar to Theorems 5, 6 and 7. It easy to see that the topology induced by the norm on the set E is stronger than the topology of the space

406

Chapter 11

Thus, developing the indicated idea about connection between generalized solutions in Banach and in locally convex spaces, we could represent the results of Subsection 1.3 in the style of Subsection 1.4 basing on the total linear subset of the set R(L ), and vice versa, we could represent the results of Subsection 1.4 in the style of Subsection 1.3 using the analogies of the a priori inequalities, i.e. assuming that the separable locally convex topology defined on the set E is weaker than the norm Finally, note that as far as many aspects of the methods used above have a topological character then the equation (1) we can investigate using the same approach in a locally convex topological spaces E and F (which can be, possibly, general topological spaces). In this case the role of is played by the completion of E by the topology induced by the systems of semi-norms

where

is a semi-norm system giving the topology of the

space F , and instead of the estimations (6) we obtain the chain of the dense continuous imbeddings where H is a completion of the set E in some locally convex topology which is weaker than the norm of the space

2. PARAMETRIZATION OF SINGULAR CONTROL IN GENERAL CASE Consider the general problem of singular optimal control of systems with distributed parameters. Let L be a linear partial differential operator (for example, pseudo-parabolic) defined in the space

407

PERSPECTIVES

is a regular tube) and having the domain of definition D(L) consisting of sufficiently smooth in

functions

obeying some uniform boundary conditions (bd ). The formally adjoint operator and its domain of definition we shall denote by and respectively, where is a set of smooth in functions satisfying the uniform adjoint boundary conditions Define on the linear manifolds

the positive norms

induced by the corresponding inner products for which the following relations hold true

and the topological condition [165] for the pairs of the neighbour norms in the inequalities is valid. By completing we obtain the chain of the positive Hilbert spaces and this imbeddings are dense and the imbedding operators are continuous. Adding to the chains spaces which are negative with respect to and are constructed by the positive norm introduced above, we obtain

Suppose that for the operators inequalities hold true

L,

the following a priori

408

Chapter 11

Then the operator can be considered as an operator continuously mapping the whole space into For the operator equation Lu = f the following results are valid: 1) for any element

there exists a unique function

such that Lu = f and the estimation valid; 2) for any element

is

there exists a unique function

such that (where

are bilinear forms constructed with the help

of extension of the inner product

onto

and onto

respectively) by continuity and the following estimation holds true Similar results are valid also for the equation with the adjoint operator Let the system state is a solution of the equation in the sense 2), where

is a mapping of the reflexive

Banach space of controls V into the negative space

is

bounded, closed, and convex subset of V. On the solutions of the equation (1) some functional is defined. The optimization problem consists in finding out such controls

on

which the performance functional attains its minimal values. The existence of the optimal controls is provided by the following assumptions: is a weakly lower semicontinuous functional; is a weakly continuous mapping.

PERSPECTIVES

409

The natural and generally accepted idea of approximate solving infinite-dimensional optimization problems consists in reducing this problem to finite-dimensional extremal problems. Consider the general principles allowing to approximate a solution of a singular optimization problem by a solution of some finite-dimensional problem. This approach we shall call parametrization of control. Suppose that V is a separable reflexive Banach space. Then there exists a sequence of finite-dimensional subspaces of the space V satisfying lim inf Next, let

the

condition

of

the

boundary

density:

be a sequence of closed arid convex subsets

which are bounded uniformly with respect to s in the norm of V and approximate the set of the admissible controls in the following sense:

The issue of the nature of the spaces, which provide the best approximation, is the most important and fine in the numerical analysis. The problem of oprimization of the system (1) we replace by the following problem:

T h e o r e m 1 . Under the conditions a), b), (2), (3) and also 1) is a strongly upper semicontinuous functional; 2) is a strongly continuous mapping; for an arbitrary has a solution

the optimal control problem (4), (5) and

Chapter 11

410

It is possible to extract from the sequence subsequence

that

such

weakly in V, where

is an optimal control of the system (1). P r o o f . Since the conditions a) and b) hold true and

are

closed and bounded subsets of the finite-dimensional subspaces

the

problem of optimal control of the system (1) and the problem (4) and (5) are solvable. Let be some optimal control of the system (1), and is a sequence of solutions of the problems (4) and (5). Then by the condition (2) there exists a sequence of controls

such that

the strong continuity of the mapping

By virtue of we have

The a priori inequality where

are generalized solutions of the state

equation of the system (1) corresponding to the controls

and

and (6) imply that

Since functional,

Consider the sequence

is a strongly upper semicontinuous

is bounded uniformly

with respect to s in the reflexive Banach space V. Thus, by the Eberlein-Shmuljan theorem there exists such subsequence that

PERSPECTIVES

411

weakly in V. The

condition

(3)

implies

that

The

mapping

is a weakly continuous (assumption b)), hence weakly in The sequence

is bounded in the norm

estimation

The

implies that

sequence bounded in

Extract from

a

is a weakly

convergent subsequence weakly in Let us prove that

Indeed, we have

Passing to the limit in (8) as

Thus,

we obtain

is a generalized solution of the system state

equation (1) corresponding to the control is unique, the sequence Let us prove that

weakly converges to is the optimal control of the system (1).

The weak lower semi-continuity of the functional (7) imply that

thus, (1). The theorem is proved.

Since this solution

and

is the optimal control of the system

Chapter 11

412

R e m a r k . If the functional is strongly continuous then lim If the optimal control of the system (1) is unique then

weakly in V.

3. DIFFERENTIAL PROPERTIES OF PERFORMANCE CRITERION IN GENERAL CASE To solve numerically the problems of simulation and optimization of systems the gradient descent methods are usually used. To use these methods correctly we must at first to investigate the smoothness of the performance functional and to prove that the necessary conditions of optimality hold true. The purpose of this section is to study the issue of the smoothness of the performance functional for the singular optimization problem in general case [169-172]. Consider the optimization problem for the system (2.1)

T h e o r e m 1 . Let the system state is defined as a solution of the equation (2.1) with the right-hand side 1) there exists a Frechet derivative at the point

of the mapping ;

2) there exists a Fréchet derivative functional

of the

at the point

then the performance functional is differentiable by Frechet at the point and the derivative is defined by the expression

413

PERSPECTIVES

where

is a solution of the adjoint problem

P r o o f. It is clear that adjoint problem

is meaningful, since the has a unique generalized solution

take on its values in mapping,

We have

In is clear that

and

Indeed, for all

is a increment of a- control. Consider

Denote

Let the problem

is a linear

is a linear functional in V, whose continuity

follows from the continuity of

Let

Since

function

is a unique solution of the problem

v be a solution We can write that

of

the

adjoint

Chapter 11

414

Applying the Schwarz inequality to the second summand in the right-hand side of (1), we have

For

the a priori estimation is valid

The assumptions of the theorem imply that for any 1)

2)

Choose

condition

such that

implies that

Then the

PERSPECTIVES

415

Thus, the theorem is proved..

T h e o r e m 2. Let 1) the mapping

has a Fréchet derivative in

the neighbour of the point the 2) the functional

,which is continuous at has a Fréchet derivative

in the neighbour of the point ,which is continuous at the point The derivative of the performance functional is continuous at the point P r o o f. Let h is an arbitrary point in the vicinity of the point Consider for all the following difference

where

is solutions of the adjoint problems:

Using the Schwarz inequality and the continuity of the linear operators

Whence,

416

Chapter 11

The derivative of the mapping since

there

is continuous at the point

exists

a

neighbour

in which

Thus,

The following a priori estimation is valid therefore The assumptions of the theorem imply that for any

1) 2)

For

the following a priori inequality holds true:

Since the mapping

has a Fréchet derivative at the point

it is continuous at the point

Choose such

that

in (2) implies that proved.

i.e.

Then the assumption The theorem is

PERSPECTIVES

417

D e f i n i t i o n 1 . A mapping from a linear normalized space X into a linear normalized space Y obeys the Lipschitz condition with index i n the set, if

T h e o r e m 3. Let 1) the mapping

has a Fréchet derivative

obeying the Lipschitz condition with index

in

the bounded convex set 2) the functional has a Fréchet derivative obeying the Lipschitz condition with index in the space Under these assumptions the derivative of the performance functional satisfies the Lipschitz condition with index in the set P r o o f is similar to the proof of the previous theorem. Let be arbitrary points in the set Consider for all difference

where

the

are solutions of the adjoint problems:

Using the Schwarz inequality and the continuity of the linear operators

418

Chapter 11

Whence,

Let us prove that the derivative

is bounded in the set

Let and h is an arbitrary point that belongs to Then the following inequalities are valid

where

is the diameter of the

set Next,

The following a priori estimations hold true fixed admissible control)

is some

PERSPECTIVES

419

Applying to

the formula of

finite increments and taking into account the fact that in

is bounded

we have

Next, we obtain Whence,

where The theorem is proved. Consider applications of these theorems to the specific singular optimization problems. Consider various performance functionals of the systems (2.1) under the assumptions of Theorem 1 for the mapping f. 1. If then the derivative of the functional

defined by the formula

and the adjoint problem has the form operator of imbedding of the space 2. For the quadratic functional Theorem 1 implies the formula:

where O is an

420

Chapter 11

3. If

then the adjoint problem

has the form where is the inverse operator for the operator I , denning the isometry between the whole space and the whole space In the case of the pseudo-parabolic system investigated in Chapter 5 the operator is defined on the Smooth functions obeying the conditions in the following way:

4. Let derivative of the functional adjoint problem has the form

where

The

is defined by the formula (4), and the

4. CONVERGENCE OF GRADIENT METHODS ANALOGIES The theorems of the previous Section allows us to construct gradient iterative methods for finding out optimal controls of systems. For this purpose at every step we must solve direct and adjoint boundary value problems, which cannot be solved exactly in practice. In addition, a singular right-hand side of the state equation are usually approximated by piecewise constant or piecewise linear functions in order to regularize problem for computer simulation. Computer computations have round-off errors. Hence, it is necessary to investigate convergence and stability of methods under the perturbations of data. Since, perturbations and round-off errors in our class of problems have an additive character, we shall consider only

PERSPECTIVES

421

the case when the right-hand side of the system state equation is perturbed and all sub-problems are solved exactly. Consider the optimal control problem for the system

where

is the set of admissible controls from the

Hilbert space of controls V . We suppose that the operator L , the functional and the mapping obey the assumptions of Sections 2 and 3, is compact in the strong topology and convex. Although in the problems of oprimization of systems with distributed parameters the strong compactness does not hold as a rule, using parametrization or regularization we can approximate the original problem in a such way, that the strong compactness holds. Consider the problem with a perturbation in the right-hand side where is one-parameter set of differentiable by Fréchet mappings, which approximate in some sense the mapping Instead of the Fréchet derivative we have its estimation Since V is a Hilbert space, and

is a linear continuous

functional in V , there exists such elements respectively. We

that shall

denote them by The methods investigated below have the following structure: the sequence of controls is constructed which satisfies the conditions and the control is a solution of some auxiliary extremal problem. The precision of

Chapter 11

422

computation of the Fréchet derivative is selected according to the conditions

where

are infinitesimal sequences of positive numbers.

To prove the convergence of these methods we shall use the sufficient conditions of the convergence of the algorithms of non-linear and stochastic programming.

4.1 Analogy of Rosen's gradient projection method Let the sequence of controls is generated by the following procedure: (i) Start from Put s = 0 . (ii) For all integer positive s compute

where s

is the number of iteration,

approximation,

is the initial

is the operator of projection in the set

is

the step multiplier selected by the condition This gradient procedure is an analogy of the well-known Rosen's gradient projection method. Consider the set U of admissible controls, for which the necessary condition of the local minimum of the functional in the set

holds true, i.e.

PERSPECTIVES

423

The necessary condition of the local minimum can be written in the form which is convenient for justification of the convergence of the method.

T h e o r e m 1 . Let 1) be a differentiable by Fréchet mapping, whose derivative obeys the Lipschitz condition with index 2)

is a differentiable by Fréchet functional, whose derivative obeys the Lipschitz condition with index If the functional takes on at the most countable number of the values in the set then all limit points (which exist necessarily) of the sequence

belong to the

compact connected subset U* and numerical sequence has a limit. P r o o f. Let us test the assumptions of the theorem of sufficient conditions of the iterative algorithms convergence. By construction, all entries of the sequence belong to the compact Consider a sequence

We have

diam Let is a subsequence convergent to the control us prove that there exists such that for all k and

Let

Chapter 11

424

Suppose the contrary. Let for all that

there exists such

for all

Then by the

triangle inequality we have

for all Put is continuous in

Here we use the notation The assumption of the theorem implies that functional, the set

the most countable, and also

is at is differentiable by Fréchet

functional and the derivative

obeys the Lipschitz condition with

index

Consider the expression

where

in the set

d = diam

is the diameter of

By the inequality (1), we shall estimate the value To do this put

Since

and write

there exists such

The following inequality holds true:

that

PERSPECTIVES

425

whence, we have

The value

we estimate in the following way

Estimate the second summand in the right-hand side of the last inequality

Next,

where problems

are solutions of the adjoint

The following estimations are valid (we use the a priori estimations of the solutions of the original and adjoint problems, the formula of

Chapter 11

426

finite increments and the fact that the derivatives

obey

the Lipschitz condition):

We have

Choosing sufficiently small

and large

we obtain

Thus, ultimately we obtain

Adding the inequalities (2) for

we obtain

PERSPECTIVES

427

Since

there For

exist

all

the is

relation

estimation valid.

The

implies the existence of such number

that

If p > max

then

Passing to the limit in (3) as condition the continuous functional Thus, there exists

and taking into account the

we arrive to contradiction with the fact that is bounded below on the compact such that for all k and

However, taking sufficiently small

and large

possible to repeat the proof of the estimation (3) for the other hand

Therefore,

it is On

428

Chapter 11

Granting the last inequality in (3), we have

Whence, Thus, the sufficient conditions of the convergence hold true, and hence, all limit points of the sequence belong to U* and the numerical sequence

has a limit. The connectedness and

compactness of the set of limit points

follows from the fact that and from the following

assertion: if the sequence of the points

of the metric space X is

imbedded into some compact set of limit points

then the

is a connected compact.

If the assumption that the set of values of the functional

is at

the most countable in the set U does not hold true, it is possible to prove weaker assertion: T h e o r e m 2 . If the assumptions 1) and 2) of Theorem 1 hold true, then the sequence has at least one limit point belonging to the set U .

4.2 Analogy of the conditional gradient method with averaging Consider the sequence of controls procedure:

generated by the following

PERSPECTIVES

429

(i) Start from Put s = 0. (ii) For all integer positive s compute

where s

is the number of iteration,

approximation,

is the initial

is the multipliers selected by the conditions

T h e o r e m 3 . Let the assumptions 1) and 2) of Theorem 1 holds true. If the functional takes on at the most countable number of its values in the set then all limit points (which exist necessarily) of the sequence

generated by

the method (i), (ii) belong to the compact connected subset U* and the numerical sequence has a limit. P r o o f is similar to the proof of Theorem 1. Let us prove only that for the sequence which converges to the control there exists

such

that

for

all

Suppose the contrary. Let for all that

for all

k

and

there exist such Then we have

Chapter 11

430

for all

Put

Consider

where

For

In the inequality (4) we estimate the value this purpose put

and write

But

Thus,

Since

there whence

exists

such

that

PERSPECTIVES

431

The value

we shall estimate in the following way

Further (in Lemma 1) it will be proved that We have Choosing sufficiently small

and large

we obtain

Thus, ultimately we have

Passing to the limit in (5) as condition

and taking into account the

we arrive to contradiction with the fact

that the continuous functional Thus, there exists

is bounded below in the compact such that for all k and

Further, proof is similar to the proof of Theorem 1. L e m m a 1 . Let the assumptions of Theorem 3 hold true. Then P r o o f. Let us introduce the following notation

Chapter 11

432

Consider

Since

obeys the Lipschitz condition with index

in the set

Let us estimate the value

Reasoning as in

the proof of Theorem 1, we have Thus, where It is clear that

Indeed,

This relation and (6) imply that If the assumption that the set of the values of the functional

in

the set U is at most countable does not hold true, then it is possible to prove the following weaker assertion.

PERSPECTIVES

433

T h e o r e m 4 . If the assumptions 1) and 2) of Theorem 1 holds true, then the sequence generated by the method (i), (ii) has at least one limit point which belongs to the set U*

5. OPTIMIZATION OF PARABOLIC SYSTEMS WITH GENERALIZED COEFFICIENTS The solvability of parabolic systems with discontinuous coefficients were investigated in [173,174,178] and in many others papers. In partucular, these problem arise as a result of investigation of heat and mass transport in heterogeneous media with non-ideal contact between subdomains, with external condensed source and so on. In this subsection the results obtained in [173,174] are extended and the issue of optimization of parabolic systems with discontinuous solutions are considered.

5.1 Main notations Let the system state described by the function The heat and mass transport take place in a tube where variables

is bounded regular domain of variation of space with boundary

Introduce the following notations. Let differentiable in the domain conditions be a completion of the set

be a set of infinitely

functions satisfying the boundary

in the norm

Chapter 11

434

be a completion (the set of smooth in

functions) in the

norm

Denote also by

a completion of the set

consisting of

functions, which are smooth in the domain and satisfy the adjoint conditions in the same norm (1). are negative spaces constructed on the basis of pairs with corresponding indices. In addition, introduce the following notations:

We shall also consider the spaces adjoint to For example,

For every Cartesian product of

the original space and its adjoint space (for example, X and X*) the bilinear form which is obtained by extension the inner product in the space by continuity, is defined Consider the problem of optimization of the system governed by the equations of heat and mass transport in several heterogeneous media with the conditions of non-ideal contact, external condensed source and so on., which generalized the similar problem considered in [174] (the Dirichlet problem)

PERSPECTIVES

435

The operator L is defined by symbolic matrix

where u is a function describing the heat and mass transport, is a vector of the substance flux. The operator L maps X into Y with the domain of definition The mapping F (possibly, non-linear) maps the Banach space of controls (with the domain of definition ) into the space Y The system coefficients obey the conditions: in the equations (2) understood to be a linear functional defined in rule

where space adjoint to the Sobolev space The functional (3) is defined correctly, since

is

by the following

is a

436

Chapter 11

We shall suppose that the coefficient matrix vector-functions

where the constant

for all

obey the conditions

does nor depend on the functions

Note that all results obtained in this section can be generalized for the functionals of the form

The oprimization problem consists in the finding out such control which provide the minimum of the performance functional

where

are some functional,

is a solution of (2) under the

control Denote by L* the adjoint by Lagrange operator

PERSPECTIVES

437

where

5.2 Optimization of parabolic system with generalized coefficients Using the technique of the a priori inequalities in negative norms [167], it is possible to prove that the following lemmas holds true. L e m m a 1 . The following inequalities are valid

where the constant c > 0 does not depend on x,y. The inequalities in Lemma 1 do not allow to extend by continuity the operator L(L ) to an operator continuously mapping the whole space X(Y ) into the space Y ( X , respectively). In what follows, we shall consider only extended operators denoting it by L è L , again. L e m m a 2 . The following inequalities hold true

For a parabolic equation with usual coefficients the a priori inequalities in negative norms were proved in [175]. The inequalities (5)-(8) allow us to prove the unique solvability of the equations (2), (4). In what follows, we shall suppose that the set is dense in the space

and also

is dense in

where

R(L), R(L ) are the ranges of values of the extended operatores (but some assertions can be proved under weaker assumptions). In the

438

Chapter 11

case when the coefficient matrix

is generated by a

classic functions the density follows from the classic theorems of solvability of parabolic equations. D e f i n i t i o n 1 . A solution of the equation (2) is an element such that for any the following equality holds true: T h e o r e m 1 . For all there exists a unique solution of the equation (2) in the sense of Theorem 1. D e f i n i t i o n 2. A solution of the equation (2) is an element such that for any the following equality holds true: T h e o r e m 2 . For all there exists a unique solution if the equation (2) in the sense of the Definition 2. The proof of Theorems 1 and 2 are similar to [167]. Remark 1 . Similar theorems hold true for the adjoint operator. Namely, for any there exists a unique solution of the equation (4) in the sense of the analogy of Definition 1 (2, respectively) for the adjoint operator. Remark 2 . If is a solution of the equation (2) with the right-hand side

in the sense of Definition 2 and

is a solution of the equation (4) with the right-hand side in the sense of the analogy of Definition 1 then Similarly, for all

the following equality holds true

PERSPECTIVES

439

where x,y are solutions of the equations (2), (4) with the right-hand sides F, G in the sense of Definitions 1 and 2, respectively. R e m a r k 3 . Note that since Theorems 1 and 2 hold true for arbitrary right-hand sides F (not only considered in [173, 174]), the results proved above are applicable not only to systems with singular coefficients, but also to systems with singular right-hand sides. For example, the righthand side

corresponds to

in the standard equation of heat and mass transport with constant coefficients. T h e o r e m 3 . Let the system state is defined from the equations (2). If 1) the functionals

are

weakly lower semicontinuous in the spaces respectively; 2) the set of the admissible controls is weakly compact in the Banach space 3)

and the mapping

continuous, then there exists a control the performance functional . P r o o f . Let us select a sequence functional ,

Granting the weak compactness of sequence

weakly converges to

is weakly providing a minimum of minimizing the

we can consider that the in the space

Taking

440

Chapter 11

into account that F is weakly continuous, we have that weakly converges to

in

Similarly to [167] we prove that

there exists such constant C > 0 (common for all

that

where x is a solution of the equation (2) with the right-hand side F , whence we conclude that the sequence of solutions of (2) under the control is bounded in and hence, we can extract from it a subsequence, which weakly converges to denote it by

to the limit as

( we

again). Passing in the equality

we obtain

whence, we conclude that is a solution of (2) under the control Taking into account that functionals are weakly lower semicontinuous we conclude that

is an optimal control.

R e m a r k 4 . A similar theorem holds true in the case when functional is weakly lower semicontinuous in the space and the mapping

is weakly continuous.

R e m a r k 5 . Assuming that

is a compact set, we can

prove that any minimizing sequence

converges to the set V*

of optimal controls R e m a r k 6 . As far as the mapping F may be non-linear and the optimization problem may be non-convex, an optimal control may be not unique.

PERSPECTIVES

If we

441

suppose

that

there

exist

the

Fréchet

derivatives

of the functionals the points

at

and also there exists the derivative

of the mapping

at the point

then we can

investigate the differential properties of the functional . T h e o r e m 4 . Let the system state be defined by the equations (2) and there exist the Fréchet derivatives Then there exists the Fréchet derivative if the functional

defined by the following expressions

where

are

the solutions of the operator equations

the function

is defined by the Fréchet derivative

by the formula (the Riesz theorem)

R e m a r k 7. A similar theorem holds true in he case of the functional and the mapping R e m a r k 8. Developing the technique proposed in [167] it is possible to investigate the smoothness of the functional gradient on the basis of the relations (9), (10) and inequalities (5)-(8) and to construct gradient type numerical methods of optimization.

REFERENCES 1. 2.

3. 4. 5. 6. 7. 8. 9. 10.

11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

Pontryagin L.S., Boltyansky V.G., Gamkrelidze R.V., Mischenko Ye.A. Mathematical Theory of Optimal Processes. Moscow, Nauka, 1969. Gamcrelidze R.V. On some extremal problems in the theory of differential equations with applications to the theory of optimal control. SIAM J. Control 1965; 3: 106—128. Feldbaum A.A. Basis of Optimal Autimatic Systems Theory. Moscow: Nauka, 1966. Bellman R.E., Dreyfus S.E. Applied Dynamic Programming. Princeton, New Jersey: Princeton University press. 1962. Tikhomirov V.M. Some issues of Approximation Theory. Moscow: Moscow Stat e University, 1976. Krasovski N.N. The Theory of Motion Control. Moscow: Nauka, 1968 (in Russian). Pschenichnyi B.N. Convex Analysis and Extremum Problems. Moscow: Nauka, 1980. (In Russian). Pschenichnyi B.N. Necessary Conditions of Extremum. New York: Marcel Decker, 1972. Mordukhovich B.S. Approximation Mathods in Problems of Optimization and Control. Moscow: Nauka, 1988. Pschenichnyi B.N., Danilin Yu.M. Numerical. Methods .in Extremum Problems. Moscow: Nauka, 1975. (In Russian.) Warga J. Optimal Control of Differential and Functional Equations. New York and London: Academic press, 1974. Bensoussan A., Lions J.L. Controle Impulsionnel et Inequations Quasivariationnelles. Paris: Dunod. Bordas, 1982. Bublik B.N., Kirichenko N.F. Foundations of Control Theory. Kiev: Vyscha shkola, 1975. Butkovski A.G. Theory of Optimal Control of Systems with Distributed Parameters. Moscow: Nauka, 1965. Butkovski A.G. Methods of Control of Systems with Distributed Parameters. Ì oskow: Nauka, 1975. Butkovski A.G., Samoylenko Yu.I. Control of Quantum-Mechanics Processes. Moscow: Nauka, 1984. Vasilyev A.G. Methods to Solve Extremal Problems. Moscow: Nauka, 1981. Egorov A.I. Optimal Control of Linear Systems. Kiev: Vyscha shkola, 1988. Egorov A.I. Optimal Control of Thermal and Diffusion Processes. Moscow: Nauka, 1978. Egorov A.I., Kapustyan V.E. "Synthesis of pulse optimal controls in distributed systems." In Control and optimization of thermal and diffusion processes, Preprint 84-34, Kiev, Inst. of Cybernetics, 1984.

444

21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38.

39.

REFERENCES

Ermoliev Yu.M. Methods of Stochastic Programming. Moscow: Nauka, 1979. Ermoliev Yu.M., GulenkoV.P., Tzarenko T.I. Finite-Difference Method in Optimal Control Problems. Kiev: Naukova Dumka, 1978. Ermoliev Yu.M., Lyashko I.I. Mikhalevich V.S., Tyuptya V.I. Mathematical Methods of Operation Research. Kiev: Vyscha shkola, 1979. Ermoliev Yu.M., Verchenko P.I. On method of linearization in limit extremal problems. Cybernetics 1976; 2:73-79. Ermoliev Yu.M., Nurminsky Ye.A. Limit extremal problems. Cybernetics 1973; 4: 130—132. Ivanenko V.I., Mel'nik V.S. Variational Methods in Control Problems of Systems with Distributed Parameters. Kiev: Naukova dumka, 1988. Lions J.-L. Controlabilite exacte des system distibutes. Comptes Rendue der Seances de L'Academic des Sciences. Ser. s. 302; 13:471—475. Lions J.-L. Controle de systemes gouvernes par des equations aux derivees partielles. Paris: Dunod Gauthier-Villars, 1968. Lions J.-L. Exact controllability, stabilization and perturbation for distributed systems. SIAM Review 1988; 1:1-68. Lions J.-L. Quelques methodes de resolution de problemes aux limities non lineaires. Dunod Gauthier-Villars Paris. 1969. Lions J.-L. Remarks on somenew non-linear boundary value problem. Lecture Notes in Nath. 1975; 446:301-328. Lions J.-L. Some remarks on free boundary problems and optimal control. In Free boundary problems. Proc. Semin. Vol. 2: Pavia, 1979. Lions J.-L. Some issues of optimal control of distributed systems. Uspekhi matem. nauk 1985; 4:55-68. Lions J.-L., Magenes E. Problemes aux limities non homogenes et applications. Paris: Dunod, 1968. Lurje K.A. Optimal Control in Problems of Mathematical Physics. Moscow: Nauka, 1975. Nakonechny A.G. Minimax estimation of functionals of solutions of variations equations in Hilbert space. Kiev: KGU, 1985. Osipov Yu.S.,Suyetov A.P. On one Lions's problem. Dokl. AN SSSR 1984; v.276, 2:288-291. Samoilenko A.M. Realization of fast-acting negative feedback on the basis of dynamic system with accumulated energy. Avtomatika i telemekhanika 1978; 12:12-23. Sirazetdinov T.K. Optimization of Systems with Distributed Parameters. Moscow: Nauka, 1977.

REFERENCES

40.

445

Fedorenko R.P. Approximate Solution of Optimal Control Problem. Moscow,

Nauka, 1978. 41. 42.

43. 44. 45. 46. 47. 48. 49. 50. 51.

52. 53. 54.

55.

Dykhta V.A., Samsonyuk O.N. Optimal impulse control with applications. Moscow: Fizmatlit, 2000. Novosel'sky S.N., Shul'gin D.F. Computation of non-stationary water transport under trickle and underground irrigation. Izvestiya of AN SSSR. Meckhanika zhidkosti i gaza 1984; 4:74-81. Obraztsov N.N. Mathematical Simulation and Optimization of Environment. Moscow: Nauka, 1985. Nents M.F. Controlling a Lethal Growth Process. Math. Biosciences 1968; 1/2: 41-55. Pavlidis T. Optimal control of pulse frequency modulated systems. JEEE Trans. 1966; 4:678-684. Chernous'ko F.L., Melikyan A.A. Game Problems of Searching and Control. Mockow: Nauka, 1978. Aslanian A.A. Necessary conditions of optimality-in problems of control with pulse impact at non-fixed time moments. Dokl. AN USSR 1982; 9:58-61. Miller B.M. Optimality condition in control problem for system described by differential equations with measure. Avtomatika i telemekhanika 1982; 6:60-73. Menaldi J.L. On the optimal stopping time problem for degenerate diffusions. SIAM J. Control 1980; 6:697-721. Orlov Yu.V. Theory of Optimal System with Generalized Control. Moscow: Nauka, 1988. Balass M.I. Sub-optimality and stability of linear distributed-parameter systems with finite-dimensional controllers. J. of Optimization Theory and Appl. 1985; 1:1-19. White L.W. Optimal bang-bang controls arising in a Sobolev impulse control problem. J. Math. Anal. and Appl. 1984; 1:237-247. White L.W. Point control of pseudo-parabolic problems. J. Differ. Equat. 1981; 3:366-374. Anisovich V.V., Kryukov B.M. Optimal control of periodic movement of linear systems with pulse impact, Prikladnaya matematika i mekhanika 1982; 2:346348. Gaskevich I.V. Investigation of some continuous-descrete problems of oscillation theory. In Controllable dynamical systems with contiunoits-descrete

parameters. Kiev: Naukova dumka, 1984. 56.

Kurzhanski A.B. Control and Observation under Conditions of Uncertainty. Moscow: Nauka, 1977.

446

57. 58. 59. 60. 61. 62. 63. 64.

65. 66. 67. 68. 69. 70. 71. 72.

73. 74.

REFERENCES

Maksudov M.V., Aliyev M.V. Optimization of pulse systems with nondistributed boundary conditions. Dokl. AN SSSR 1985; 4:796-798. Miller B.M. Non-linear pulse control of objects described by ordinary differential equations. Avtomatika i telemekhanika 1978; 1:75-86. Samoilenko A.M., Perestyuk N.A. Differential Equations with Impulse Impact. Kiev: Vischa shkola, 1980. Tzypkin Ya.Z., Popkov Ya.C. Theory of Non-Linear Impulse Systems. Moscow: Nauka, 1973. Blaquire A. Impulsive optimal control with finite or infinite time horizons. J. Optimiz. Theory and Appl. 1985; 4:431—439. Didenko V.P. Variational method of solving boundary value problems with nonsimmetric operator. Dokl. AN SSSR 1978; 6:1277-1280. Didenko V.P. On boundary problems for multidimensional hyperbolic boundare value problems with singularity. Dokl. AN SSSR 1972; 4:763-766. Didenko V.P., Popova A.A. "Inequalities in. negative norms and variational problems for non-simmetric differential iperators." Preprint 81-29, Kiev, Inst. of Cybernetics, 1981. Komkov V. Optimal Control Theory for the Damping of Vibrations of Simple Elastic Systems. Berlin, Heidelberg, New York:Springer-Verlag. 1972. Krotov V.F., Gurman V.I. Methods and Problems of Optimal Control Problems. Moscow: Nauka, 1973. Samborsky S.N. On controllability of systems described by partial differential. Sibirsky matem. zhurnal 1983; 2:150-159. Silin D.B. Discontinuous optimal controls in linear problems. Dokl. AN SSSR 1985;2:314-316. Goodson R.E., Klein R.E. A definition and some results for distributed system observability. JEEE Trans. on aut. control 1970; 2:165-174. Meirovich L., Silverberg L.M. Control of non-selfadjoint distributed-parameter systems, J. of Optimization Theory and Appl. 1985; 1:77-90. Sokolowski J. On optimal parametric control of parabolic control of parabolic systems. Lect. Notes Comp. Sci. 1976; 41:623-633. Teo K.L. Convergence of a strong variational algorithm for relaxed controls involving a distributed optimal control problem of parabolic type. Numer. Funct. Anal. and Optim. 1984-1985; 2-3:125-144. Teo K.L., Reid D.W. First-order variation algorithm for optimal problems involving parabolic systems. Numer. Func. Anal. and Ottim. 1982; 2:141-171. Berezanski Yu.M. Expansion of Self-Adjoint Operators by Eigenfunctions. Kiev: Naukova Dumka, 1965.

REFERENCES

75. 76. 77. 78. 79. 80. 81. 82. 83.

84. 85. 86. 87. 88. 89.

90. 91. 92. 93.

447

Krein S.G., Petunin Yu.I. Scales of Banach spaces. Uspekhi Matem. Nauk 1966; 2:89-169. Banach S.S. Course of Functional Analysis. Kiev: Radyanska shkola, 1948. Kantorovitch L.V., Akilov G.P. Functional Analisys. Moscow: Nauka, 1977. Yosida K. Functional Analysis. Berlin, Gottingen, Heidelberg: Springer-Verlag, 1965. Belov Yu.A., Didenko V.P., Kozlov N.N. et al. Mathematical Software of Complex Experiment. Vol.1. Kiev: Naukova Dumka, 1982. Lyashko S.I. Pulse control of linear systems with distributed parameters. Dokl. An SSSR 1984; v.276, 2:285-287. Lyashko S.I. Approximate solving pulse optimal control problems. Dokl. AN USSR 1986; 2:60-62. Lyashko S.I. Numerical methods of solving pulse optimal control problems J. Vychislitel'noy Matematiki i Matematicheskoy Physiki 1987; 3:470-475. Lyashko S.I., Man’kovsky A.A. Simultaneous optimization of impulse moments and intensities in control problems for parabolic equations. Cybernetics 1983; 5:687-690. Lyashko S.I., Man’kovsky A.A. Controllability of parabolic systems with impulse action. Soviet Math. Dokl. 1989; v.39, 3:476-479. Lyashko S.I., Red’ko S.E. Optimal pulse-point control of the dynamics of a viscous stratified fluid. Differentsial'nye Uravneniya 1987; 11:1890-1897. Lyashko S.I., Red’ko S.E. On the controllability of distributed systems in a class of generalized actions. Kibernetika 1988; 6:120-121. Chuprun B.Ye. Problems with constraints on changing of vector control. Avtomatika i telemekhanika 1976; 12:20-31. Lavrentyev M.M., Romanov V.G., Shishatsky S.P. Non-correct Problems of Mathematical Physics and Analysis. Moscow: Nauka, 1980. Lyashko I.I., Didenko V.P. On Tikhonov type regularization of problems with symmetric operator in a Hilbert space. Dokl. AN SSSR 1976; v.226, 3: 510— 512. Morozov V.A. Regular methods of solving non-correctly posed problems. Preprint of Computational Center of Moscow State University, 1974. Prilepko A.I. Inverse problems of potential theory. Elliptic, parabolic, hyperbolic equations and transport equation. Mat. zametki 1973; 5:757-767. Tikhonov A.N., Arsenin V.Ya. Methods of Solving Non-Correct Problems. Moscow: Nauka, 1979. Sobolev S.L. Some Applications of Functional Analysis in Mathematical Physics. Leningrad: Leningrad State University, 1950.

448

94. 95. 96. 97. 98. 99.

100. 101. 102.

103.

104.

105. 106.

107. 108.

REFERENCES

Lyashko S.I. On regularization of pulse-point control in pseudo-parabolic systems. Kibernetika i vychiclitel'naya tekhnika 1986; 71:25-27. Gupal A.M. Stochactic Methods of Solving Nonsmooth Extremal Problems. Kiev: Naukova dumka, 1979. Karmanov V.G. Mathematical Programming. Moscow: Nauka, 1986. Katkovnik V.Ya. Linear Estimations and Stochastic Optimization Problems. Moscow: Nauka, 1976. Levitin Ye.S., Polyak B.T. Methods of Minimization with Constraints. J. Vychislitel'noy Matematiki i Matematicheskoy Physiki, 1966; 5:787-823. Moiseev N.N. Mathematical Problems of System Analysis. Moscow: Nauka, 1981. Shor N.Z. Methods of Minimization of Non-Differentiable Functions and Their Applications. Kiev: Naukova dumka, 1979. Gaivoronski A.A., Ermoliev Yu.M. Stochatic programming problems with sequential estimating. Izv. AN SSSR, Techninal cybernetics 1979; 4:81-83. Lyashko S.I., Golodnikov A.N. Method of minimizing functionals with respect to distrbution functions and its application in probability theory Cybernetics 1979; 4:532-536. Mikhalevich V.S., Gupal A.M., Bolgarin I.V. Numerical methods of solving non-stationary and limit extremal problems without computation of gradient. Kibernetika, 1984; 5:56-57. Zoutendijk G. Methods of Feasible Directions. A Study in Linear and NonLinear Programming. Amsterdam, London, New York, Princeton: Elsevier publishing company, 1960. Moiseev N.N. Mathematical Methods in Optimal Control Systems. Moscow: Nauka, 1971. Ekeland I., Temam R. Convex Analysis and Variational Problems. NorthHolland publishing company. Amsterdam, Oxford. American elsevier publishing company, inc. New York. 1976. Nurminsky Ye.A. Numerical methods of solving determinate and stochastic minimax problems. Kiev: Naukova dumka, 1979. Lyashko S.I. The method of feasible directions in the solution of exrtemal problems with unknown parameters. In Mathematical methods in operations research and reability theory. Akad. Nauk Ukrain. SSR, Inst. Kibernet., Kiev, 1978. (In Russian)

REFERENCES

449

109. Lyashko S.I. A method of linearization in the solution of exrtemal problems for functions with unknown parameters. In Mathematical methods in operations research and reability theory. Akad. Nauk Ukrain. SSR, Inst. Kibernet., Kiev, 1978. (In Russian.) 110. Lyashko S.I., Mirzoahmedov F. Some methods of simultaneous identification and optimization. Dokl. Tadjikskoy AN SSR. 1984; 12:721-724. 111. Rubinstein L.M. On issue about heat transport in heterogeneous media. Izvestiyta AN SSSR. Ser. georgaph. i geophys. 1948; 1: 27-46. 112. Barenblatt G.I. On some boundary values problems for equations of liquid filtration in fissured rock. Prikladnaya matematika i mekhanika 1963; 2:348-350 113. Barenblatt G.I., Entov V.M., Ryzhik V.M. Theory of Non-stationary Filtration of Liquid and Gas. Moscow: Nauka, 1978. 114. Barenblatt G.I., Zheltov Yu.P., Kochina I.N. On basic idea of theory of homogeneous liquid filtration in fissured rocks. Prikladnaya matematika i mekhanika, 1960; 5:852-864. 115. Dzektzer Ye.S. Generalization of equation of grounwater mowement with free surface. Dokl. AN SSSR 1972; 5:27-46. 116. Nao R.C., Ting T.W. Initial value problems for pseudoparabolic differential equations. Indian uniq. Math. J. 1973; 25:348-368. 117. Belova G.P, Bodrov V.A., Globus A.M. et al. Hydrophysical properties of ions and mathematical simulations of salt movement in soil. Pochvovedeniye 1984; 4:113-119. 118. Lyubonova A.Sh. Some Issues of Boundary Problems Theory for PseudoParabolic Equations. Krasnoyarsk, 1989. 119. Atamanov E.R., Mamayusupov M.Sh. Non-classical problems for pseudoparabolic equations. Frunze: Ilim, 1990. 120. Lyashko S.I. Some issues of pulse-point control in pseudo-parabolic systems. Ukr. Matem. J. 1985; 3:368-371. 121. Lyashko S.I. Pulse-point control of pseudo-parabolic systems. Cybernetics 1986; 2:122-123. 122. Lyashko S.I. On solvability of pseudo-parabolic equations. Izv. vysshykh uchebnykh zavedeniy, Matematika 1985; 9:71—72. 123. Malovichko V.O. Non-local boundary value problem for singular pseudoparabolic system. Doklady AN USSR, Ser. À. 1990; 3:25-27. 124. Matveev N.M., Svirdyuk G.A. On one linear pseudo-parabolic equation of nonstationary filtration. Mathematical physics 1985: 276.

450

REFERENCES

125. Pavlov D.A., Potapov M.M. On iniform boundenness of solutions of distributed model of intra-cell mobility. Vestnik Moskovckogo Universiteta. Ser. 15, Vichisl. matem. i kibern. 1995; 2:62-64. 126. Suveyka I.V. On Cauchy problem for non-stationary equations with small viscosity. Mat. invest. 1988; 79:89-100. 127. Suveyka I.V. Mixed problems for non-stationary equations. Mat. invest. 1980; 59:27-36. 128. Showalter R.E., Ting T.W. Pseudo-parabolic partial differential equations. SIAM T. Math. Anal. 1970; 1:1-26. 129. Garding L. Cauchy's Problem for Hyperbolic Equations. University of Chicago. 1957. 130. Il'in V.A. On solvability of mixed problems for hyperbolic and parabolic equations. Uspekhi Matem. Nauk 1960; 2:97-154. 131. Ladyzhenskaya O.A. Boundary Value Problems of Mathematical Physics. Moscow: Nauka, 1973. 132. Lyashko S.I., Man’kovsky A.A. Optimal impulse control of a system with distributed parameters of hyperbolic type. Dokl. AN USSR Ser. A 1983; 4: 6971. 133. Tikhonov A.N., Samarsky A.A. Equations of Mathematical Physics. Moscow: Nauka, 1977. 134. Plumb O.A., Whitaker S. Dispersion in heterogeneous porous media. Water Resour. Res. 1988; 7:913-938. 135. Shemiakin E.I. Propagation of non-stationary perturbations in viscoelastic medium. Dokl. AN SSSR 1955; 1:34-37. 136. Kachanov L.M. Foundations of Plasticity Theory. Moscow: Nauka, 1981. 137. Ismatov M. About one mixed problem for equation describing the sound propagation in viscous gas. Differentzialniye uravneniya 1984; 6:. 1023-1024. 138. Malovischko V.A. On boundary value problems for singular pseudo-parabolic and pseudo-hyperbolic systems. Differentzialniye uravneniya 1991; 12:21202124. 139. Suveyka I.V. Mixed problems for equations of perturbations propagation in viscous nedia. Differentsizlniye Uravneniya 1983; 2:337-347. 140. Kozhanov A.I. Problem with directional derivative for some pseudo-parabolic and similar equations. Sibirian mathematical journal 1996; 6: 1335-1346. 141. Krein S.G. Linear Differential equations in a Banach space. Moscow: Nauka, 1967. 142. Lyashko I.I., Didenko V.P., Tzitritzky O.Ye. Filtration of Noises. Kiev: Naukova dumka, 1979.

REFERENCES

451

143. Lyashko S.I. Method to solve approximately the Cauchy problem with operator coefficient of hyperbolic type. Dokl. AN SSSR 1979; v.247, 3:546549. 144. Lyashko S.I. Approximate solving control problems with operator coefficient of hyperbolic type. Preprint 82-34, Kiev, Inst. of Cybernetics, 1982. 145. Kato T. Theory-of Linear Operators Perturbations. Moscow: Mir, 1972. 146. Vishik M.I. Cauchy problem for equations with operator coefficients, mixed boundary value problem for systém of differential equations and approximate method of its solving. Matematichesky sbornik 1956; 1:51-148. 147. Sobolev S.L. About one new problem of mathematical physics. Izvestiya An SSSR, Saer. Mat. 1954; 1:3-50. 148. Zeleniak T.I. On invariant subspaces of one operator. Sibir. Matem. J. 1962; 3:471-475. of solutions of one S.L.Sobolev's 149. Demidenko G.V. Estimations as problem. Sibir. Matem. J. 1984; 2:112-120. 150. Skazka V.V. Asimptotics as of solutions of one problem of mathematical physics. Matem. sbornik 1985; 1:3-40. 151. Gabov S.A., Malysheva G.O. On Cauchy problem for one class of movements of viscous stratified liquid. J. Vychislitel'noy Matematiki i Matematicheskoy Physiki 1984; 3:467-471. 152. Gabov S.A., Malysheva G.O., Sveshnikov A.G. About one equation of dynamics of viscous stratified liquid. Differentsial'nye Uravneniya 1984; 7:1156-1164. 153. Louis J.-C1., Wexler D. On exact controllability in Hilbert spaces. Lect. Notes in Cotr. and Inf. Sei. 1984; 58:676-685. 154. Niane M.T. Controlability exacte de 1'equation des plaques vibrantes dans un polygone. Comptes Rendus de L'Academie des sciences 1988; tome 307, Serie 1, 10:517-521. 155. Astrovski A.I., Gayshun I.V. Controllability of linear non-stationary systems in clas s of finite order distributions. Izvestiya AN. Teoriya i sistemy upravleniya 1998; 2:24-30. 156. Semenov V.V. Impulse controllability of some linear distributed systems of Sobolev type. Dopovidi NAN Ukrayiny (to appear). 157. Lyashko S.L, Semenov V.V. On controllability of linear distributed systems in classes of generalized impacts. Kibernetika i sistemny analiz 2001; 1:18-41. 158. Berezansky Yu.M. Self-Adjoint Operators in Spaces of Functions of Infinite Variable Number. Kiev: Naukova Dumka, 1987.

452

REFERENCES

159. Ladyzhenskaya O.A., Ural'tzeva N.N. Linear and Quasi-linear Elliptic Equations. Moscow: Nauka, 1964. 160. Garkusha V.I. On one approach to solving equation of first kind with simmmetric operator of Fredhol type. Ukr. Matem. J. 1979; 6:666-674. 161. Lyashko S.I., Red’ko S.E. An approximate solution for a problem of the dynamics of a viscous stratified fluid. J. Vychislitel'noy Matematiki i Matematicheskoy Physiki 1987; 5:720-729. 162. Lyashko S.I., Red’ko S.E. Approximate solution of equation of viscous stratified liquid dynamics. Dokl. AN SSSR 1986; v.290, 1:15-17. 163. Sakawa Y. Controllability for partial differential equations of parabolic type. SIAM Journal on Control 1977; 3:383-400. 164. Toshihiro K. Discrete-time controllability for distributed-parameter systems. J. L. System Sciences 1980; 9:1063-1074. 165. Klyushin D.A., Kuschan A.A., Lyashko S:I, Nomirovsky D.A., Petunin Yu.IG eneralized sopution of some operator equations in Banach spaces. J. Obchyslyuvalnoyi i Pryladnoyi Matematiki 2001 (to appear). 166. Hille E., Phillips R. Functional Analysis and Semigroups. Revised edition. American mathematical society colloquivem publications, v. XXXI, Providence. 1957. 167. Lyashko S.I. Generalized Control of Linear Systems. Kiev: Naukova dumka, 1998. 168. Petunin Yu.I., Plichko A.N. Theory of Characteristics of Subspaces and its application. Kiev: Vischa shkola, 1980. 169. Semenov V.V. Differentiability of performance criterion in problems of generalized optimal control of distributed systems. Dopovidi NAN Ukrayiny 2000; 10:111-114. 170. Semenov V.V., Bubnov V.V. Derivative of performance criterion in problems of generalized optimal control of linear distributed systems J. Obchyslyuval'noyi i prykladnoyi matematyky 1999; 1(84): 129-136. 171. Nomirovsky D.A On convergence of gradient type methods in solving singular optimal control problems. J. Obchyslyaval'noyi ta prykladnoyi matematyky 1998; l(83):93-98. 172. Nomirovsky D.A. Control in pseudo-hyperbolic systems. Obchyslyuval'na ta prykladna matematyka 1996; 80:68-77. 173. Deyneka V.S., Sergienko I.V., Skopetzky V.V. Mathematical Models and Methods to Solve Problems with Discontinuous Solutions. Kiev: Naukova Dumka, 1995.

REFERENCES

453

174. Lyashko I.I., Demchenko L.I., Mistetzky G.Ye. Numerical Solution of Heat and Mass Transport Problems in Porous Media. Kiev: Naukova dumka, 1991. 175. Berdimuratov M.K., Lyashko S.I., Nomirovsky D.A. Simulation and optimization of the pickling processes of the irrigated area. Obchyslyuval'na ta prykladna matematyka 1996; 80:3-11. 176. Nomirovsky D.A. Approximate methods of the control of the pseudo-hyperbolic systems. Vesnik KSU. 1997, 3: Ñ. 187-198. 177. Lyashko I.I., Lyashko S.I., Nomirovsky D.A. Controllabilaty of hyperbolic and pseudo-hyperbolic systems by the singular impacts. Dopovidi NAN Ukrayiny 2000; 11: 131-134. 178. Nomirovsky D.A. Optimization of the parabolic systems with generalized coefficients. Dopovidi NAN Ukrayiny 2000; 12: 77-82.

Applied Optimization 36. 37.

G. Di Pillo and F. Giannessi (eds.): Nonlinear Optimization and Related Topics. 2000 ISBN 0-7923-6109-1 V. Tsurkov: Hierarchical Optimization and Mathematical Physics. 2000 ISBN 0-7923-6175-X

38. 39. 40. 41. 42. 43. 44.

C. Zopounidis and M. Doumpos: Intelligent Decision Aiding Systems Based on Multiple Criteria for Financial Engineering. 2000 ISBN 0-7923-6273-X X. Yang, A.I. Mees, M. Fisher and L.Jennings (eds.): Progress in Optimization. Contributions from Australasia. 2000 ISBN 0-7923-6286-1 D. Butnariu and A.N. Iusem: Totally Convex Functions for Fixed Points Computation and Infinite Dimensional Optimization. 2000 ISBN 0-7923-6287-X J. Mockus: A Set of Examples of Global and Discrete Optimization. Applications of Bayesian Heuristic Approach. 2000 ISBN 0-7923-6359-0 H. Neunzert and A.H. Siddiqi: Topics in Industrial Mathematics. Case Studies and Related Mathematical Methods. 2000 ISBN 0-7923-6417-1 K. Kogan and E. Khmelnitsky: Scheduling: Control-Based Theory and PolynomialTime Algorithms. 2000 ISBN 0-7923-6486-4 E. Triantaphyllou: Multi-Criteria Decision Making Methods. A Comparative Study. 2000

ISBN 0-7923-6607-7

45.

S.H. Zanakis, G. Doukidis and C. Zopounidis (eds.): Decision Making: Recent Developments and Worldwide Applications. 2000 ISBN 0-7923-6621-2

46.

G.E. Stavroulakis: Inverse and Crack Identification Problems in Engineering Mechanics. 2000 ISBN 0-7923-6690-5 A. Rubinov and B. Glover (eds.): Optimization and Related Topics. 2001 ISBN 0-7923-6732-4

47. 48.

M. Pursula and J. Niittymäki (eds.): Mathematical Methods on Optimization in Transportation Systems. 2000 ISBN 0-7923-6774-X

49.

E. Cascetta: Transportation Systems Engineering: Theory and Methods. 2001 ISBN 0-7923-6792-8

M.C. Ferris, O.L. Mangasarian and J.-S. Pang (eds.): Complementarity: Applications, Algorithms and Extensions. 2001 ISBN 0-7923-6816-9 51. V. Tsurkov: Large-scale Optimization – Problems and Methods. 2001 ISBN 0-7923-6817-7 52. X. Yang, K.L. Teo and L. Caccetta (eds.): Optimization Methods and Applications. 2001 ISBN 0-7923-6866-5 53. S.M. Stefanov: Separable Programming Theory and Methods. 2001 ISBN 0-7923-6882-7 50.

Applied Optimization 54.

S.P. U ryasev and P.M. Pardalos. (eds.): Stochastic Optimization: Algorithms and Applications. 2001 ISBN 0792369513

55.

J. G ilAluja (ed.): Handbook of Management under Uncertainty. 2001 ISBN 0792370252

56.

B.N. Vo, A. Cantoni and K.L. Teo: Filter Design with Time Domain Mask Con straints: Theory and Applications. 2001 ISBN 0792371380

57. 58.

S. Zlobec: Stable Parametric Programming. 2001

ISBN 0792371399

M.G. N icholls, S. Clarke and B. Lehaney (eds.): MixedMode Modelling: Mixing Methodologies for Organisational Intervention. 2001 ISBN 0792371518

59.

F. G iannessi, P.M. Pardalos and T. Rapcsák (eds.): Optimization Theory. Recent Developments from Mátraháza. 2001 ISBN 140200009X

60.

K.M. H angos, R. Lakner and M. G erzson: Intelligent Control Systems. An Introduc tion with Examples. 2001 ISBN 1402001347

61. 62.

D. G stach: Estimating OutputSpecific Efficiencies. 2002

ISBN 1402004834

J. G eunes, P.M. Pardalos and H.E. Romeijn (eds.): Supply Chain Management: Models, Applications, and Research Directions. 2002 ISBN 1402004877

63.

M. G endreau and P. Marcotte (eds.): Transportation and Network Analysis: Current Trends. Miscellanea in Honor of Michael Florian. 2002 ISBN 1402004885

64.

M. Patriksson and M. Labbé (eds.): Transportation Planning. State of the Art. 2002 ISBN 1402005466

65.

E. de Klerk: Aspects of Semidefinite Programming. Interior Point Algorithms and Selected Applications. 2002 ISBN 1402005474

66.

R. Murphey and P.M. Pardalos (eds.): Cooperative Control and Optimization. 2002 ISBN 1402005490

67.

R. Corrêa, I. D utra, M. F iallos and F. G omes (eds.): Models for Parallel and Distri buted Computation. Theory, Algorithmic Techniques and Applications. 2002 ISBN 1402006233

68.

G . Cristescu and L. Lupşa: NonConnected Convexities and Applications. 2002 ISBN 1402006241

69.

S.I. Lyashko: Generalized Optimal Control of Linear Systems with Distributed Para meters. 2002 ISBN 140200625X

KLUWER ACADEMIC PUBLISHERS – DORDRECHT / BOSTON / LONDON

Generalized Optimal Control of Linear Systems with Distributed Parameters (Applied Optimization)

Applied Optimal Control: Optimization, Estimation and Control

Applied Optimal Control: Optimization, Estimation, And Control

Optimal control of nonsmooth distributed parameter systems

Optimization and Optimal Control

Linear Optimal Control

Linear Systems: Optimal and Robust Control

Linear systems: optimal and robust control

Some Aspects of the Optimal Control of Distributed Parameter Systems

Some Aspects of the optimal Control of distributed parameter systems

Some Aspects Of The Optimal Control Of Distributed Parameter Systems

Some Aspects of the Optimal Control of Distributed Parameter Systems

Optimal Control Systems

Optimal Adaptive Control Systems

Introduction to applied optimal control

Intelligent Control Systems: An Introduction with Examples (Applied Optimization, Volume 60) (Applied Optimization)

Control of Continuous Linear Systems

Cooperative Control and Optimization (Applied Optimization)

Almost Optimal Adaptive Control Systems

Generalized Linear Models: An Applied Approach

Optimal control with engineering applications

Generalized linear models. An applied approach

Variational calculus and optimal control: Optimization with elementary convexity

Applied Linear Optimal Control:Examples and Algorithms

Applied Linear Optimal Control:Examples and Algorithms

Optimal Control of Nonlinear Parabolic Systems

Cooperative Control of Distributed Multi-Agent Systems

Dynamics and Control of Distributed Systems

Optimal Control

Cooperative systems: control and optimization

Generalized Optimal Control of Linear Systems with Distributed Parameters (Applied Optimization)

Applied Optimal Control: Optimization, Estimation and Control

Applied Optimal Control: Optimization, Estimation, And Control

Optimal control of nonsmooth distributed parameter systems

Optimization and Optimal Control

Linear Optimal Control

Linear Systems: Optimal and Robust Control

Linear systems: optimal and robust control

Some Aspects of the Optimal Control of Distributed Parameter Systems

Some Aspects of the optimal Control of distributed parameter systems

Some Aspects Of The Optimal Control Of Distributed Parameter Systems

Some Aspects of the Optimal Control of Distributed Parameter Systems

Optimal Control Systems

Optimal Adaptive Control Systems

Introduction to applied optimal control

Intelligent Control Systems: An Introduction with Examples (Applied Optimization, Volume 60) (Applied Optimization)

Control of Continuous Linear Systems

Cooperative Control and Optimization (Applied Optimization)

Almost Optimal Adaptive Control Systems

Generalized Linear Models: An Applied Approach

Optimal control with engineering applications

Generalized linear models. An applied approach

Variational calculus and optimal control: Optimization with elementary convexity

Applied Linear Optimal Control:Examples and Algorithms

Applied Linear Optimal Control:Examples and Algorithms

Optimal Control of Nonlinear Parabolic Systems

Cooperative Control of Distributed Multi-Agent Systems

Dynamics and Control of Distributed Systems

Optimal Control

Cooperative systems: control and optimization

Recommend Documents